# Keep a character, person or style the same across many images

> Keep a character, person or style the same across many images: 0 companies fully do it, 12 with a limit, 45 carry the tag. Every listing in Image generation with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/image-generation.character-consistency/
- Markdown: https://www.anchorterminal.com/niches/image-generation.character-consistency/index.md (~1,400 tokens)
- Slim: https://www.anchorterminal.com/niches/image-generation.character-consistency/index.min.md (~380 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/image-generation.character-consistency/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 0 | 12 | 45 | 6 | 3 |

## What counts as done

The same subject kept across a set of images, using reference images or a fine-tune the API accepts, with the same model still callable next month.

Measured, for our own tests: across 10 images a reviewer recognises the same subject in at least 9.

How agents ask for it: same character on every page; headshots that look like me; keep the style of my earlier cards; use these references

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Adobe Firefly API (Adobe Inc.)**: Reference blobs (512 to 2048 pixels, 10 MB) and Firefly Custom Models give two routes, but no identity-preservation claim or retirement policy was found.
- **Black Forest Labs FLUX API (Black Forest Labs Inc.)**: Up to ten references on FLUX 3 and eight on FLUX.2, a character-consistency guide and finetune endpoints give two routes, but the guide is vendor examples with no recognition rate, and FLUX 3 can't yet be pinned to a dated version.
- **Bria API (Bria Artificial Intelligence Ltd.)**: Edits take up to four reference images, but no identity-preservation claim or fine-tune route was found, and there is no deprecation policy to keep a model callable.
- **fal image models (Features & Labels, Inc.)**: Multi-reference edit models and LoRA training endpoints exist, but acc-28 found only 1 or 2 of 40 portraits recognisable after eight flux-lora-fast-training runs, and deprecated endpoints carry no removal date.
- **FLORA (FLORA AND FAUNA FLOWERSHOP, INC.)**: Ideogram Character and LoRA Style training through the API give reference and fine-tune routes, but no recognition rate or model retirement policy is published.
- **Ideogram API (Ideogram AI, Inc.)**: ideogram-3-character takes character_reference_images with an optional mask, and custom-model endpoints exist, but it is the 3.0 line, no recognition rate is given, and models change without a dated changelog.
- **Leonardo.Ai API (Leonardo Interactive Pty Ltd)**: Character and style reference guidances exist on Lucid models, but no reference count or identity claim was read, and the deprecations page shows models removed with little or no notice.
- **OpenAI Image API (OpenAI OpCo, LLC)**: Edits take several reference images at high fidelity and image models get three months' notice, but the guide says recurring characters may not stay consistent, there is no fine-tune for image models, and acc-08 lost continuity across pages.
- **Qwen-Image 2.1 (Model Studio) (Alibaba Cloud (Singapore) Private Limited)**: Up to 10 reference images and identity preservation for people are model-card claims with no measured rate, and the API's model lifecycle wasn't readable.
- **Replicate image models (Replicate, LLC)**: The docs index lists a FLUX fine-tune guide and versioned LoRA models, an identity route that can be pinned by version, but no recognition rate is given and no retirement notice applies to official models.
- **Seedream on BytePlus ModelArk (BytePlus Pte. Ltd.)**: References plus generated images can total 15 in a sequential request, which suits a set of related images, but there is no fine-tune route or identity claim, and models carry dated version ids that get deactivated.
- **Stability AI Image API (Stability AI Ltd)**: Style guide and style transfer take a reference image, but there is no identity reference, no fine-tune in the API and no retirement notice period.

- not this job: Iris-3B (Speridlabs): No reference-image input and no fine-tune guide in the card or README.
- not this job: Recraft API (Recraft Inc.): Reference images (1 to 10) define a style, not a subject, and no fine-tune or identity reference is offered in the API.
- not this job: SurgePix API (SurgePix): No reference-image input or fine-tune is documented for generation skills.

## What stops agents finishing

- model that defined the style is retired
- fine-tune preserves clothing and lighting but not identity
- continuity lost over long multi-image projects

## Around it

Before: storage.download. After: video-generation.image-to-video, image-generation.project-asset-batch.

## Where this comes from

Weight 6 of 100 in Image generation: 3 of 37 accounts in this category (game style lost to deprecation, manga continuity, LoRA headshots at 1 or 2 of 40); 11 video-generation accounts show the same need upstream of image-to-video

As data: https://www.anchorterminal.com/api/v1/use-cases/image-generation.character-consistency.json, or the MCP tool get_use_case with id image-generation.character-consistency. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
