creative

Best AI Character Generators

The best AI character generators, ranked on independent evidence: how well each one designs an original character and keeps it the same across images, with benchmark results, prices, free tiers, and what you can own and what you cannot generate.

Updated October 2026

Quick answer: The best AI character generator for most people is Google Gemini with its Nano Banana image models, because it is free to use, Google says it keeps up to five characters consistent from reference images, and Nano Banana 2 scored highest on identity preservation in TRACE-Bench, the most recent independent multi-reference benchmark. For the best-looking single character design, ChatGPT is the better choice, because its GPT Image 2.5 models rank first and second on the Artificial Analysis text-to-image and editing leaderboards. For a stylised, art-directed look, Midjourney is the pick, and for a dedicated character workspace with model training, Leonardo.ai is. It was a close call between Gemini and ChatGPT, and one caveat applies to every tool here: no model holds a character perfectly, and the best score on TRACE-Bench’s identity test was 0.77 out of 1.

This page ranks tools for designing an original character (a story protagonist, a game or tabletop character, a mascot, an OC) and keeping that character the same across many images. That is a different job from the pages next to it:

• Best AI avatar generators covers a likeness of yourself: talking-head video avatars, profile pictures and VTuber models.
• Best AI comic generators covers the panel workflow: layout, speech bubbles, lettering and page-by-page production.
• Best AI image generators covers which model makes the best single image of anything.
• This page covers the character itself: designing someone who does not exist, producing a character sheet, and getting the same face, outfit and proportions back in the next scene.

If you searched “AI character generator” because you want a character to chat with rather than a picture, see the section on character chatbots below.

Which one should you pick?

What the evidence says about AI character consistency

Character consistency is the whole difficulty in this category. Designing one good-looking character is easy for every tool here; getting the same character back in a second scene is not. Four independent sources measure it, and they agree on the shape of the answer.

Reference images work better than descriptions. Every tool on this page is more consistent when you give it an image of the character than when you describe the character again in words. IMG.LY’s character consistency benchmark described one character in precise detail and asked 23 models for the same person in three scenes, with no reference image, and found that “outfits change, faces shift, props disappear” between scenes. Scores ranged from 4.86 out of 5 for Seedream 5.0 Pro down to 2.50 for Recraft V3. IMG.LY sells a creative-editing SDK and promotes its own editor on the same page, so treat it as a useful test rather than a neutral one.

Nano Banana 2 holds identity best from references, in the one recent benchmark that tests it. TRACE-Bench, published on 17 August 2026 by Haoran Wang, Chaofan Ma, Ran Yi and Lizhuang Ma of Shanghai Jiao Tong University, scores multi-reference generation on about 1,600 cases built from around 4,000 reference images. On its Anchor test, which measures whether an entity keeps its identity, Nano Banana 2 scored 0.7724 and GPT-Image-1.5 scored 0.7649; across all models tested, Anchor scores ran from 0.52 to 0.77. GPT Image 2.5 was released after the paper and was not tested.

More characters means less consistency. The MultiBanana benchmark, accepted at CVPR 2026, found that scores fall as references are added: GPT-Image-1 scored 7.80 out of 10 with one reference and 6.59 with two, and the original Nano Banana fell from 7.82 to 4.89. Both are older models than the current ones, but the direction matters for anyone putting two original characters in the same scene.

Overall image quality is led by OpenAI. On the Artificial Analysis text-to-image leaderboard, GPT Image 2.5 Sunburst leads at 1197 Elo and GPT Image 2.5 Flare is second at 1191, with Nano Banana 2 sixth at 1125 and Nano Banana Pro eleventh at 1102. On the image editing leaderboard, which tests changes to an existing image and is closer to the “same character, new scene” job, GPT Image 2.5 Sunburst leads at 1180 and Nano Banana 2 is eighth at 1108.

Independent results at a glance

The models in each test differ, and so do the methods, so read across a row with care. Midjourney’s current V8.2 is not on the Artificial Analysis board, and the V7 Alpha figure there should not be taken as a measure of V8.2.

How we judged these tools

A tool is ranked here only when there is independent evidence about it from more than one kind of source: benchmarks and arena leaderboards, research papers, expert reviews, and user ratings with real volume behind them. Vendors’ own claims about consistency, user counts and quality are on the page, labelled as their claims, never as the reason for a rank.

The number one is the tool that does the most character jobs well for the most people (designing a character, holding it across scenes, putting two characters together), with free access breaking ties. Tools that are credible but lack independent assessment are listed separately under “Other character tools worth knowing (not ranked)”.

Best AI character generators ranked

Prices are each vendor’s published US list price.

The best AI character generators, reviewed

1. Google Gemini with Nano Banana — best AI character generator overall

Models: Nano Banana 2 (Gemini 3.1 Flash Image), the default; Nano Banana Pro (Gemini 3 Pro Image), for subscribers Price: Free in the Gemini app with limited daily quotas; Google AI Pro $19.99 a month for higher limits Consistency: Up to 14 reference images; resemblance of up to 5 characters, according to Google Watermark: Visible watermark on free and Pro images; SynthID invisible watermark on all

Gemini is the best AI character generator for most people because it is free, it is built to take references, and it has the best independent identity score. Google launched Nano Banana 2 on 26 February 2026 and says it keeps “character resemblance of up to five characters and the fidelity of up to 14 objects in a single workflow”. Nano Banana Pro, released on 20 November 2025, makes the same five-person claim from up to 14 input images. Nano Banana 2 is the free model in the Gemini app; subscribers can switch to Pro. Google says free users get “limited free quotas” and then fall back to an older model; it does not publish a fixed daily number.

The independent evidence is the broadest of any tool here. Nano Banana 2 had the highest identity-preservation score in TRACE-Bench (0.7724), it sits sixth on the Artificial Analysis text-to-image leaderboard at 1125 Elo, and it scored 3.75 out of 5 in IMG.LY’s prompt-only test. That last score is mid-table, which is the point: Gemini holds a character well when you give it a reference, and less well when you only describe one.

Why it wins: Free access, explicit multi-character consistency from references, and the top reference-based identity score of any model tested.

Limitations: OpenAI’s GPT Image 2.5 beats it on overall image quality in blind voting. Google does not publish free limits. Since a Disney cease-and-desist letter in December 2025, Gemini refuses to generate Disney characters, which matters only if you were trying to make someone else’s character.

Best for: Anyone designing an original character from scratch and wanting it back, consistently, in new scenes, without paying.

2. ChatGPT with GPT Image 2.5 — best for character image quality

Models: GPT Image 2.5, in two variants (Flare, the fast default, and Sunburst, the more detailed); ChatGPT chooses automatically Price: Free with limits; Plus $20 a month Consistency: Reference images in conversation; edits build on earlier edits Released: 8 September 2026

ChatGPT is the best choice when the quality of the character design matters most. OpenAI released ChatGPT Images 2.5 on 8 September 2026 to all ChatGPT users, free tier included. OpenAI says image subjects “look more recognizable” and “distinctive features are more likely to carry through” from reference photos, and that “earlier changes are more likely to stay consistent” across a run of edits. OpenAI does not claim that a character will stay the same across separate generations.

The independent evidence puts it first on quality. GPT Image 2.5 Sunburst and Flare hold first and second place on both the text-to-image and editing leaderboards at Artificial Analysis. In IMG.LY’s prompt-only consistency test, Sunburst scored 4.72 out of 5, second only to Seedream 5.0 Pro. Its predecessor, GPT-Image-1.5, scored 0.7649 on TRACE-Bench’s identity test, just behind Nano Banana 2.

Why it ranks second: It is the stronger image model, but Gemini publishes a multi-character consistency spec and has the better reference-based identity result, and both are free. That was the deciding margin, and it was close.

Limitations: OpenAI publishes no image limit for free users or Plus. You cannot choose between Flare and Sunburst in ChatGPT. Recurring characters across separate chats rely on you re-attaching the reference.

Best for: Designing a character’s look from a written description, and refining it by conversation until it is right.

3. Midjourney — best for art-directed character design

Version: V8.2 Price: Basic $10 a month, Standard $30, Pro $60, Mega $120; about 20% off billed annually; no free plan Consistency: The V8.2 edit model generates from up to 4 image references at once

Midjourney is the best choice for characters that need a distinctive, painterly or cinematic look. On 27 August 2026 Midjourney released an edit model for V8.2 that edits images from instructions and generates “with other images (replacing omni-reference) with up to 4 image references at once”. That replaces Omni Reference (--oref), the character feature most guides still describe, which now works only on the older V7. Midjourney’s announcement makes no specific character-consistency claim, and it says style references “may work best with some extra prompt direction in this version”.

The independent evidence is expert opinion rather than measurement. Midjourney does not offer a public API, so its current version is not on the Artificial Analysis leaderboard; the only entry there is V7 Alpha, at 878 Elo and 113th, which is not a measure of V8.2. Our own image generator page and most expert round-ups rate its aesthetic as the strongest for art. Midjourney is also a defendant in the copyright cases brought by Disney, Universal and Warner Bros. Discovery, covered below.

Why it ranks third: The best-looking stylised characters, but no free plan and no independent consistency measurement.

Limitations: No free plan. Four references at most. Guides and tutorials written before August 2026 describe --cref and --oref, which do not work on V8.2.

Best for: Concept artists, writers and game designers who care most about the look of a character and will art-direct it.

4. Leonardo.ai — best dedicated character workspace

Price: Free, 150 tokens a day; Essential $12 a month; Premium $30; Ultimate $60; up to 20% off billed annually Consistency: One or several reference images; Element training for a custom character model; Multi-Character Scene Builder for up to 5 characters Models: Its own plus partner models, including Nano Banana and GPT Image Owner: Canva

Leonardo.ai is the best choice for people who want character design as a workflow rather than a chat. Its character generator takes “one or multiple reference images” to “lock in your character’s core look”, offers Element Training for a personalised model, and has a Multi-Character Scene Builder that will “combine up to five characters into one scene”. It runs partner models including Nano Banana and GPT Image alongside its own, and an auto model switcher picks one for you. Leonardo says commercial rights apply “even on the free plan”, subject to its terms. Its pricing page lists Essential at $12 a month with 8,500 tokens and 10 model trainings a month, Premium at $30 and Ultimate at $60. Canva acquired Leonardo when it had more than 19 million users, according to Canva.

The independent evidence is an expert review and a large body of user ratings. TechRadar’s review recommended Leonardo for quality, value and speed and criticised a clunky interface, but it dates from July 2024 and does not test character consistency. On Trustpilot, Leonardo scores 4.7 out of 5 from 1,487 reviews, with most complaints about billing. Zapier’s 2026 round-up of AI art generators includes it as a tool aimed at professional creatives.

Why it ranks fourth: The most complete set of character tools in one place, including access to the two top-ranked models, but no independent test of its own consistency features.

Limitations: The token system makes cost hard to predict, and partner models cost more tokens. Free-plan images are public. The most detailed vendor documentation of Character Reference dates from 2024 and covers older models.

Best for: Game developers, writers and hobbyists building a cast of characters who want training, references and multi-character scenes in one subscription.

5. Ideogram Character — best free consistency from one image

Price: Free, weekly slow credits; Plus $20 a month or $15 a month billed annually; Pro $60 a month or $42 billed annually Consistency: A single reference image; face and hair auto-masked; works with Remix and Magic Fill Launched: 29 July 2025

Ideogram Character is the best free option when you have one image of your character and want more. Ideogram launched it on 29 July 2025, free for all users on the web and iOS, and describes it as working “with a single input photo” for “characters (both real and imagined)”. Ideogram’s documentation says it masks the face and hair of the reference automatically. Ideogram’s models are also known for rendering text inside images, which suits character name cards and posters. Ideogram’s pricing page lists Plus at $20 a month, or $15 a month billed annually, with 1,000 priority credits a month.

The independent evidence is leaderboard and benchmark placement plus expert coverage. Ideogram 4.5 (High) sits 34th on the Artificial Analysis text-to-image leaderboard at 1013 Elo; Ideogram 3.0 scored 3.75 out of 5 in IMG.LY’s consistency test; and Zapier lists its character creator among its picks.

Why it ranks fifth: The simplest free route from one image to a consistent character, but well below the top models on image quality.

Limitations: One reference image only. Ideogram does not say how many credits a Character generation costs or which model version it uses. Free credits arrive weekly and generate slowly.

Best for: Someone with a single drawing or render of a character who wants it in new poses and scenes for free.

6. Seedream 5.0 Pro — best for consistency without reference images

Maker: ByteDance Released: 8 July 2026 Where to use it: CapCut, Dreamina and the BytePlus API Price: API $0.075 per image (Pro), $0.035 (Lite); consumer pricing in Dreamina data not available Consistency: Up to 10 reference images

Seedream 5.0 Pro is the most consistent model when you describe a character rather than show it. It scored 4.86 out of 5 in IMG.LY’s prompt-only consistency test, the highest of 23 models, and it sits 15th on the Artificial Analysis text-to-image leaderboard at 1081 Elo and joint ninth on editing at 1107. According to CapCut, it takes up to 10 reference images and costs $0.075 per image by API. A third-party listing claims 14 references on another platform; CapCut’s figure of 10 is used here.

Why it ranks sixth: The best independent consistency score without references, but reached mainly through ByteDance’s video and editing apps or an API, and its consumer pricing could not be confirmed.

Limitations: No standalone app of its own. IMG.LY’s test is a single vendor-run benchmark. Seedream does not lead on reference-based identity, which is the more common workflow.

Best for: CapCut and Dreamina users, and developers building character features through an API.

7. FLUX.2 — best for running a character model yourself

Maker: Black Forest Labs Versions: max, pro and flex (API); dev and klein (open weights) Price: API pro $0.03 per megapixel, max $0.07 per megapixel; open-weight versions free to run locally Consistency: Up to 8 reference images by API and 10 in the playground on max, pro and flex; 4 on klein

FLUX.2 is the best choice for people who want to run a character model on their own hardware or build it into a product. Black Forest Labs’ documentation shows eight consistent characters generated from reference images and lists reference limits of 8 by API on max, pro and flex, and 4 on klein. FLUX.2 [klein] 4B is released under the Apache 2.0 licence; licence terms for the other open-weight versions differ, so check them before commercial use.

The independent evidence is leaderboard and benchmark placement. FLUX.2 [flex] is 22nd on the Artificial Analysis text-to-image leaderboard at 1028 Elo, and FLUX.2 scored 3.75 out of 5 in IMG.LY’s consistency test.

Why it ranks seventh: The strongest option for local and custom use, and the only open-weight model here, but behind the closed models on quality and with no consumer app of its own.

Limitations: No first-party consumer app. Running it locally needs a capable GPU and some setup. Fine-tuning a character (a LoRA) is a separate, technical step.

Best for: Developers, studios that need to keep assets in-house, and hobbyists who already run image models locally.

How to design a character with AI and keep it consistent

The research above and the vendors’ own documentation point to the same five-step workflow, whichever tool you use.

1. Design the character before you need it in a scene. Generate a front-facing image on a plain background first, and iterate until the face, hair, build and outfit are right. Every tool here holds a character better from a clean reference than from a busy scene.

2. Make a character sheet. Ask for the same character from the front, side and back, in a neutral pose, plus a row of expressions. This becomes your reference set. Krea, Leonardo and Higgsfield all offer character-sheet or turnaround features, and any general model will make one on request.

3. Always attach the reference. IMG.LY’s test found that a detailed written description does not reliably bring back the same character. Re-attach the sheet in every new chat or project, in Gemini, ChatGPT, Midjourney and Ideogram alike.

4. Add characters one at a time. MultiBanana found consistency falls as references are added. Get each character right alone, then combine them; Gemini claims up to five characters and Leonardo’s scene builder up to five, but expect more drift with each.

5. Train a model for a long project. For a character who will appear in hundreds of images, a trained model (Leonardo’s Element Training, Scenario, Krea or a FLUX.2 LoRA) holds the design more tightly than references. The trade-off is set-up time and less flexibility if the character changes. Our comic generators page compares these consistency methods for panel work.

Pricing comparison

The cheapest complete set-up costs nothing. Design the character and make a character sheet in Gemini or ChatGPT, then use the sheet as a reference in either app, or in Ideogram Character, for every new scene. Paying makes sense for volume (Google AI Pro or ChatGPT Plus), for a style (Midjourney), or for training a character model (Leonardo).

Best AI character generator for each job

Best free AI character generator

Google Gemini is the best free AI character generator, because Nano Banana 2 is free in the Gemini app, takes reference images and is built to hold up to five characters.

Best for original characters from a text description

ChatGPT is the best for designing a character from words alone, because GPT Image 2.5 leads the Artificial Analysis text-to-image leaderboard and scored 4.72 out of 5 in IMG.LY’s prompt-only consistency test.

Best for fantasy, D&D and tabletop characters

Midjourney is the best for fantasy and tabletop character art, because its V8.2 aesthetic is the strongest for painterly and cinematic illustration. For a free alternative, use Gemini or ChatGPT. If you are making art for a publisher, check its AI policy first: Wizards of the Coast has barred its commissioned artists from using generative AI since 19 December 2023.

Best for game characters

Leonardo.ai is the best for game characters among the ranked tools, because it combines reference images, custom model training and multi-character scenes. Scenario, under the tools not ranked below, is built specifically for game studios.

Best for keeping two or more characters consistent

Google Gemini with Nano Banana Pro or Nano Banana 2 is the best for scenes with several original characters, because Google claims resemblance for up to five characters from up to 14 references. Leonardo’s Multi-Character Scene Builder also takes up to five.

Best for consistency from one image

Ideogram Character is the best for building on a single image, because it was designed to work from one reference and is free on every plan.

Best for mascots and brand characters

ChatGPT is the best for designing a mascot, because GPT Image 2.5 leads on prompt accuracy and editing. For a brand that needs the mascot rendered the same way every time, train a model: Leonardo.ai offers training on every paid plan. For logos built around a mascot, see best AI logo generators.

Best for running locally

FLUX.2 is the best for running a character model on your own computer, because it is the only model here with open weights and multi-reference support.

Best for a character based on your own face

That is an avatar, not an original character. See best AI avatar generators for talking avatars and stylised profile pictures, and best AI face swap apps for putting a face into an existing image.

Best for writing the character’s story

Image tools design the look; they do not write the backstory. See best AI story generators for tools that write the character’s history, voice and plot.

Other character tools worth knowing (not ranked)

These are credible, working products that people use to make characters, but we found too little independent assessment of their character features to rank them. Descriptions and prices are the vendors’ own unless stated.

• Higgsfield offers Soul Cast, which builds fictional characters from genre, era, archetype, physique and outfit and generates a backstory and character sheet, and Soul ID, which trains on 20 or more photos of a real face for 25 credits.
• Scenario is a game-asset platform with custom character training from 5 to 30 images and a New Character Generator that produces new characters in the style of existing ones. It has a free plan of 50 credits a day for personal use; Starter is $15 a month, or $10 a month billed annually, and training starts on Pro at $45 a month.
• Krea generates character sheets, turnarounds and expression grids, with consistency from LoRA training on up to 50 images on the free plan. Its free plan gives 100 compute units a day; Pro is $35 a month, or $21 a month billed annually.
• OpenArt released Character Builder to all users on 27 March 2026. OpenArt says it currently supports only photorealistic human characters, with anime, fantasy and 3D styles planned. It scores 3.6 out of 5 from 1,524 reviews on Trustpilot.
• Adobe Firefly opened Firefly Custom Models, which train on your own images of a character or style and are private by default, as a public beta on 19 March 2026. Firefly Standard is $9.99 a month. Adobe’s IP indemnity applies to enterprise customers who buy that entitlement, not to Standard or Pro plans.
• NightCafe lets you save a character from one to four reference images and call it by name in a prompt, and the feature works on its free daily credits.
• Perchance AI Character Generator is free, needs no sign-up and has no daily limit, according to the site.
• Artbreeder blends and edits images through its Splicer and Composer tools, with plans from $7.49 a month on its pricing page. It has no dedicated consistent-character feature.

Can you generate an existing character?

Generating someone else’s character is where AI character tools meet copyright law, and the rules have tightened since 2025.

Disney’s characters are blocked in Gemini. Disney sent Google a cease-and-desist letter on 11 December 2025, and by 11 February 2026 Gemini and Nano Banana were refusing to generate Disney characters, citing “concerns from third-party content providers”, Android Authority reported. Characters from other studios were still generated at that point.

Three studios are suing Midjourney. Disney and Universal sued Midjourney in June 2025 (C.D. Cal. No. 2:25-cv-05275) and Warner Bros. Discovery followed on 5 September 2025; the cases were consolidated in November 2025. Midjourney argues fair use and says users are responsible for what they generate. According to AI Lawsuit Tracker, summary judgment motions are due on 23 November 2026 and a trial would come in 2027 at the earliest; there has been no ruling and no settlement.

The licensed route closed with Sora. Disney agreed on 11 December 2025 to license more than 200 Disney, Marvel, Pixar and Star Wars characters to OpenAI for Sora and ChatGPT Images, alongside a $1 billion investment. After OpenAI said on 24 March 2026 that it was discontinuing Sora, Disney left the partnership, Engadget reported. See best AI video generators for what replaced Sora.

Character chatbots were affected too. Disney sent Character.AI a cease-and-desist letter at the end of September 2025, and Character.AI removed the characters, Axios reported.

The practical rule: an original character is what these tools are for. Fan art made with AI for personal use is a grey area, and selling it is a legal risk.

Who owns a character made with AI?

In the United States, images made entirely by AI are not protected by copyright. The US Copyright Office cancelled protection for the Midjourney images in the comic Zarya of the Dawn on 21 February 2023 because they “are not the product of human authorship” (US Copyright Office). Its January 2025 report on copyrightability said that “prompts alone do not provide sufficient human control”. The courts agree: the DC Circuit held in Thaler v. Perlmutter that “human authorship is a bedrock requirement”, and the Supreme Court declined to hear the appeal on 2 March 2026.

What you add can be protected. The Copyright Office says human creative input that is visible in the output, such as your own drawing used as the starting image, and creative selection, arrangement or modification of AI output, can be protected. A character’s written backstory, personality and story are your writing. Allen v. Perlmutter, a case asking whether more than 600 prompts plus editing count as authorship, is awaiting a ruling.

Elsewhere, the position is moving. The UK keeps a provision protecting computer-generated works, but the government’s 18 March 2026 report on copyright and AI said it is under review “with a preference to remove the provision” (IPKat). In the EU, Article 50 of the AI Act has applied since 2 August 2026 and requires AI-generated images to be marked as such in machine-readable form (Goodwin).

This is general information, not legal advice. If a character is central to a business, speak to an IP lawyer about copyright and trademark.

Looking for a character to chat with?

Many people who search for an “AI character generator” want a character they can talk to, not a picture. That is a companion or roleplay app, and it is covered on best AI companions. The largest of them, Character.AI, ended open-ended chat for users under 18 on 25 November 2025 (Windows Report) and has since added visual and story features, including Imagine Gallery on 18 March 2026, according to its own blog. The image tools on this page can make the portrait for a character you then bring to a chat app.

Turning a character into a 3D model

To use a character in a game engine, a 3D print or an animation, you need a 3D model, which is a separate tool, ranked on best AI 3D model generators. Two are worth knowing. Tripo auto-rigs biped and quadruped models in a Mixamo-compatible format, with a free plan of 200 credits and Pro at $20 a month for 3,000 credits; its free-plan outputs carry licence restrictions, so use a paid plan for commercial characters. Meshy has a free plan of 100 credits a month and says rigging and animation cost no credits; its free outputs are licensed CC BY 4.0, which requires credit. Start from a clean front-facing image or character sheet made with any tool above.

Frequently asked questions

What is the best AI character generator?

Google Gemini, with its Nano Banana image models, is the best AI character generator for most people, because it is free, Google says it holds up to five characters consistent from reference images, and Nano Banana 2 scored highest on identity preservation in the TRACE-Bench benchmark published in August 2026. ChatGPT with GPT Image 2.5 is the better choice for the highest image quality, Midjourney for stylised art, and Leonardo.ai for a dedicated character workspace with model training.

What is the best free AI character generator?

Google Gemini is the best free AI character generator, because Nano Banana 2 is free in the Gemini app and takes reference images to keep a character consistent. ChatGPT’s GPT Image 2.5 is also available to free users with limits. Ideogram Character is free on every Ideogram plan and works from a single reference image. Leonardo.ai gives 150 free tokens a day.

How do I make the same character in different images with AI?

Make a character sheet first, showing the same character from the front, side and back on a plain background, then attach it as a reference image every time you generate a new scene. Independent testing by IMG.LY found that describing a character again in words does not reliably reproduce it. Gemini, ChatGPT, Midjourney V8.2, Leonardo.ai, Ideogram Character, Seedream and FLUX.2 all accept reference images. For hundreds of images of one character, train a custom model in Leonardo.ai, Scenario, Krea or FLUX.2.

Can ChatGPT create consistent characters?

Yes, with limits. ChatGPT Images 2.5, released on 8 September 2026, carries distinctive features through from reference images more reliably than earlier versions, and edits build on earlier edits, according to OpenAI. OpenAI does not claim that a character stays identical across separate generations, so re-attach a reference image in each new chat. GPT Image 2.5 leads the Artificial Analysis text-to-image and editing leaderboards.

Is Gemini or ChatGPT better for character design?

ChatGPT makes the higher-quality images, ranking first on both Artificial Analysis image leaderboards against sixth and eighth for Gemini’s Nano Banana 2. Gemini is better at holding a character from references, with the top identity score in TRACE-Bench, and Google claims consistency for up to five characters at once. For designing one character’s look, use ChatGPT; for a recurring cast across many scenes, use Gemini.

Can I use Midjourney character reference on V8?

No. The --cref and --oref parameters do not work on Midjourney V8.2. On 27 August 2026 Midjourney released an edit model for V8.2 that generates from up to four image references at once and replaces Omni Reference. Omni Reference still works if you switch back to V7.

Do I own the copyright to an AI-generated character?

Not to images made entirely by AI, in the United States. The US Copyright Office has held since 2023 that purely AI-generated images lack human authorship, and the Supreme Court declined to hear the Thaler v. Perlmutter appeal on 2 March 2026. Your own contributions can be protected: a drawing you used as the starting image, substantial edits, and the character’s written backstory. This is general information, not legal advice.

Can AI generate Disney or anime characters?

Major tools increasingly refuse. Gemini stopped generating Disney characters by February 2026 after a Disney cease-and-desist letter, and Disney, Universal and Warner Bros. Discovery are suing Midjourney over generated images of their characters, with no ruling yet. Generating an existing character for personal use is a legal grey area, and selling it is a legal risk. Original characters in an anime style are fine to generate.

What is the difference between an AI character generator and an AI avatar generator?

An AI character generator designs someone who does not exist, such as a story protagonist, a game character or a mascot, and keeps that character consistent across images. An AI avatar generator makes a likeness of you or another real person, usually a talking video presenter or a stylised profile picture. Our best AI avatar generators page covers the second job, including the consent rules that apply to real faces.

What is the best AI character generator for D&D?

Midjourney is the best for D&D and tabletop character portraits, because its V8.2 aesthetic is the strongest for painterly fantasy art, from $10 a month. Google Gemini and ChatGPT are the best free options. If the art is for a commercial publisher, check its policy first, because Wizards of the Coast has barred its commissioned artists from using generative AI since December 2023.

Is Perchance AI character generator good?

Perchance’s AI character generator is free, needs no account and has no daily limit, according to the site, which makes it a quick way to try ideas. For a character you will reuse, Gemini or Ideogram Character, both free, hold the design from a reference image.

Can I make an AI character I can talk to?

Yes, but that is a companion or roleplay app rather than an image generator. Our best AI companions page compares those apps. Character.AI, the largest, has limited open-ended chat to users aged 18 and over since 25 November 2025. You can design the character’s portrait with any tool on this page and use it in a chat app.

How to choose

Start with Google Gemini if you are designing an original character and want it back in new scenes: it is free, it takes references, and it has the best independent identity score. Use ChatGPT when the look of the character matters most, or alongside Gemini, since both are free. Pay for Midjourney if you want a stylised, art-directed character and will work at its look. Choose Leonardo.ai if you are building a cast for a game or a long story and want training and multi-character scenes in one place. Use Ideogram Character if you already have one image of your character, and FLUX.2 if you need to run everything yourself.

Whatever you use, make a character sheet first and attach it every time. Every independent test on this page found that describing a character again is not enough to get the same character back.