MiniMax H3 Image character swap: what it is, what it needs, and what you can do here instead
MiniMax H3 Image character swap is a community-made Hugging Face model that swaps one person in a picture for a reference character; not from MiniMax, it needs ~140 GB of VRAM locally; this site does not run it, and its editor changes one photo in about 37 seconds for 10 credits. Free with sign-up: 10 credits = 1 image, then from $0.13 per image on a monthly plan.

Your photoStyle changes (cartoon, 3D, anime) are not reliable yet. If the result looks unchanged, try again or describe the change more specifically.
One-tap edits
- Not affiliated with MiniMax or Hugging Face
- The editor here is a different model
- Free with sign-up: 10 credits = 1 image
- VRAM for a full local run, per the model card
- ≈ 140 GB
- size of the ComfyUI checkpoint file (bf16)
- 40 GB
- author’s own comparisons that preferred the merge (one labeler)
- 14 of 17
- left out of the licence territory (LICENSE text)
- EU, UK, KR, US
What is the MiniMax H3 Image character swap?
It is a model repository on Hugging Face, not a website or app, called akhaliq/MiniMax-H3-Image-Character-Swap. The card says it fuses a character-swap LoRA made by Akatz Labs, originally trained for video, into MiniMax’s open H3 Image model, and calls it the first merged checkpoint of that kind. You give it a scene image as Picture 1 and a character reference as Picture 2, and it is meant to swap the named person while keeping the background, lighting, other people and pose. That is the author’s description; we did not run it, so we cannot say how well it works.
- A model repo with a 40 GB ComfyUI checkpoint, LoRA files, a merge script and a workflow file
- Scene image plus character reference plus a short instruction
- Author’s claim: 14 of the author’s own 17 comparisons preferred the merge, judged by one labeler


Who made it, and is it from MiniMax?
No. The repo belongs to the Hugging Face user akhaliq, whose profile says only “Building on HF”; it does not say the author works for Hugging Face or MiniMax, so we do not say so either. The base model is MiniMaxAI’s H3 Image, and the LoRA is by Akatz Labs. We found no MiniMax statement endorsing this merge. The repo has no creation date on its page, but its oldest discussion thread is about five days old, so it is probably from around November 14, which would be after the base model’s November 11 release; that is our inference, not a stated date.
- Base model: MiniMaxAI/MiniMax-H3-Image (MiniMax)
- LoRA: akatz-ai/MiniMax-H3-Character-Swap-LoRA (Akatz Labs), trained for video
- Merge and workflow: akhaliq (community, not MiniMax)

What does a character swap need to run, and who may use it?
The card says a full local run with diffusers needs about 140 GB of VRAM, because the image model alone is 66 GB and its text encoder 67 GB, and that running the text encoder on another machine leaves only about 2 GB for the local card. For ComfyUI it offers one 40 GB checkpoint and a workflow file, and says ComfyUI 0.45.0 or newer is needed; it gives no VRAM figure for that route. The licence is MiniMax’s H3 Image community licence: it is worldwide except the EU, the UK, South Korea and the United States, and written permission from MiniMax is needed above one million US dollars of yearly revenue or to offer the model as a hosted service. We read the licence text on November 19; it is not legal advice.
- About 140 GB VRAM for a full local diffusers run (card)
- 40 GB checkpoint for ComfyUI; VRAM for that route not stated
- Licence territory leaves out the EU, UK, South Korea and the US
- Consent: MiniMax’s acceptable use policy bans impersonating a person without consent


Can I do a character swap on this site?
No, not a real one. This site’s editor takes one photo and a written request of up to 400 characters, not a second reference picture, and it does not run H3 Image. Its request filter refuses face-swap wording, deepfakes and look-alikes of real or famous people, and the fixed server template refuses to make someone look like a real, named person. What it does do is the nearest honest thing: swap the outfit, the background or the art style of your own photo, or replace the person with an invented character. A typical edit takes about 37 seconds and 10 credits, at 1024 × 1024.
- One photo in, one edited photo out; no second reference picture
- Outfit, background or style swaps, or an invented character
- Refuses face swaps and look-alikes of real people
- No video: the Akatz LoRA was built for video, this site makes still images only

What we read, and what we could not confirm
The model card (the author’s own claims, not tested by us)
- A merged checkpoint: the character-swap LoRA is fused into the H3 Image diffusion transformer. Scene as Picture 1, character reference as Picture 2; it should keep background, lighting, other people and pose. Distilled: 6 steps, CFG 1. Recommended canvas about 4 megapixels (2720 × 1536); the model is native at about 16 megapixels.Source: Hugging Face model card ·
- Run with diffusers 0.43.0, or in ComfyUI with a 40 GB pre-merged checkpoint and a workflow file built on the official H3 Image edit template. The card needs ComfyUI 0.45.0 and template pack 0.11.104 or newer. The text encoder, VAE and optional prompt expander come from the Comfy-Org repo, not from this repo.Source: Hugging Face model card ·
- Author’s test: 17 comparisons (6 scenes, 2 characters, 3 seeds) judged by one labeler preferred the merge 14 times, the plain LoRA twice and the native model once. A separate set of 8 hard cases is published. This is a small self-run test, not an independent benchmark.Source: Hugging Face model card ·
- Limits the card lists: the LoRA was trained for video, the image version is checked on 8 hard cases only, swaps follow the source person’s pose and gaze, and extreme poses, heavy occlusion and very small faces can drift. It also warns about consent and likeness rights for real people.Source: Hugging Face model card ·
- Hosting: no Inference Provider runs it. The card lists six Spaces that use it. We opened one (hugging-apps): it shows “Running on Zero” and no usage text, and we did not run it.Source: Hugging Face model and Space pages ·
Other pages we read
- The LoRA’s own card describes a video-to-video LoRA for MiniMax’s H3 video model, calls it experimental, and says identity, timing, expressions, occlusions and cuts can drift. It carries no release date and no consent warning.Source: Akatz Labs LoRA card ·
- The licence text applies worldwide excluding the EU, the UK, the Republic of Korea and the United States, is dated November 10, 2026, needs MiniMax’s prior written authorization above 1 million USD of annualized revenue or for hosted services, and its acceptable use policy bans impersonating another person without that person’s consent. It has no separate deepfake clause.Source: MiniMaxAI LICENSE ·
- Comfy-Org repackages MiniMax’s H3 Image files for ComfyUI under the same community licence and links official Text to Image and Image Edit templates. It states no ComfyUI version and no hardware needs.Source: Comfy-Org on Hugging Face ·
- ComfyUI’s own docs describe H3 Image as a cloud partner node: it runs on MiniMax’s servers through a Comfy API account, with no model download and no local GPU, from ComfyUI 0.45.1. The page states no price.Source: ComfyUI docs ·
- RunComfy, a third party, lists a hosted ComfyUI workflow that uses the merged checkpoint: scene and character pictures plus a short instruction, denoise near 1.0 for a full swap, best between 4 and 20 megapixels. Its page names no GPU tier, price or consent note.Source: RunComfy (third party) ·
- The repo’s Community tab had four open threads and none closed: requests for an INT8 version, a Spaces demo and a video LoRA, plus one test report. No out-of-memory, safety or bug reports. Dates are relative only.Source: Hugging Face discussions ·
Staying safe with downloaded models and workflows
- Loading a pickle file can run code. Hugging Face advises loading models only from users and organisations you trust, and says its scanner is not foolproof and that checking is the user’s responsibility.Source: Hugging Face docs (pickle) ·
- Safetensors is described as a format for storing tensors safely, as opposed to pickle. The repo lists its checkpoint and LoRA files as .safetensors; it also ships a Python merge script, which we did not open.Source: Hugging Face docs (safetensors) and model card ·
- Custom nodes are code. Comfy Org’s January 2025 post names two past custom-node incidents, ComfyUI_LLMVISION and ultralytics, and says it plans a way to remotely disable bad nodes.Source: Comfy Org blog ·
Our own checks
- We ran the three prompts in the cards below through this site’s own request filter, in English and Chinese: all pass, at 253, 241 and 292 characters of the 400 allowed. We did not generate them, so no result is shown.Source: This site’s request filter, tested ·
- On 2026-11-18 we ran a one-off 7-photo smoke test of H3 Image editing through a third-party API (not a character swap): faces were kept in all five photos of people, and one cat’s identity was lost. It is not a likeness test, and this site does not offer H3 Image.Source: This site’s own test notes ·
Not confirmed
- The repo’s creation or update date. The page shows none, only relative dates on its threads.Source: Hugging Face discussions ·
- Whether akhaliq acts for Hugging Face or MiniMax. His profile says only “Building on HF” and shows no staff badge.Source: Hugging Face profile ·
- How much VRAM and time the ComfyUI route needs, any independent quality test, what the six Spaces require, and the RunComfy price. We also did not open the merge script or the workflow file.Source: Hugging Face model card ·
Compare
Ways to get a character swap, and what this site does
| Route | Run the merged model yourself | Hosted workflow (RunComfy, third party) | This site (photo editor) |
|---|---|---|---|
| What you give it | A scene image, a character reference picture and a short instruction | The same three inputs | One photo and a written request of up to 400 characters; no second picture |
| Hardware or setup | About 140 GB VRAM for a full diffusers run; ComfyUI route: a 40 GB checkpoint, VRAM not stated | Runs on the host’s side; GPU tier not stated on the page | Nothing to install |
| Cost | The files are downloadable; MiniMax’s written permission is needed above 1 million USD revenue or for hosted services | Not read | Free with sign-up: 10 credits = 1 image; then 10 credits per image, from $0.13 on a monthly plan |
| Licence limits | MiniMax H3 Image community licence; EU, UK, South Korea and US left out of its territory | Not stated on the page; the models keep their own licences | This site’s own terms; no H3 model involved |
| Puts a different person into my photo? | Yes, that is its purpose (the author’s claim) | Yes, per the page | No: it refuses face swaps and look-alikes of real people; it can swap clothes, background or style, or draw an invented character |
| Output size | About 4 MP recommended canvas, native about 16 MP | Scene scaled to about 4 MP; best at 4 to 20 MP | 1024 × 1024 |
Before you try it
How to try a character swap without regrets
Get consent first
Use only pictures of yourself or of people who agreed. MiniMax’s acceptable use policy bans impersonating someone without consent, and the model card warns about likeness rights. Never swap in a celebrity or a real person who did not agree.
Check the licence and your hardware
Read MiniMax’s licence for your country and use: the EU, the UK, South Korea and the US are outside its territory, and hosted services or revenue above 1 million USD need written permission. Then check the numbers: about 140 GB of VRAM for the diffusers route, a 40 GB file for ComfyUI.
Download carefully, or use a hosted route
Hugging Face’s docs say pickle files can run code when loaded and advise loading models only from users and organisations you trust; its scanner is not foolproof. The repo’s checkpoints are listed as safetensors, a format built to hold only tensors. Start from ComfyUI’s official H3 Image template, add custom nodes only from sources you trust (Comfy Org has named two past custom-node incidents), and read the merge script before you run it. We have not read it, and we found no report of a problem, which proves nothing either way.
Or try the nearest thing here
Upload one photo, paste one of the prompts below, and swap the outfit, the background or the style. It takes about 37 seconds and 10 credits. Check the face and hands before you share it.
Prompts
Swap-style prompts for this site’s editor (not run)
These three prompts are our own wording for this site’s photo editor, which runs a different model from H3 Image. We did not generate them for this page, so there is no result to show; we only checked on 2026-11-19 that they pass the editor’s request filter. They keep the face unless the card says otherwise, name no brand or person, and we have not tested them in other tools. Results vary with the photo.
Swap the outfit (not run)
More edits in the AI photo editorNot run: we spent no credits on this page. Checked only against this site’s request filter on 2026-11-19. Written for the fictional portrait shown on this page.
Swap the outfit
Keep my face, hair, pose and the park background exactly the same. Change only my clothes: swap the green sweater for a plain navy trench coat over a white shirt, same fit and natural folds. Soft daylight, realistic fabric. No text, logos or watermarks.
Swap the background (not run)
MiniMax H3 Image: what it is and what it costsNot run. Our own earlier studio-background edit (the picture at the top) is a different prompt for the same kind of change.
Swap the background
Keep my face, hair and clothes exactly the same. Swap the park for a quiet snowy street at dusk, warm shop windows glowing and soft out-of-focus lights behind me, and match the light on my face to the new scene. No text, logos or watermarks.
Swap me for an invented character (not run)
The ComfyUI workflow files for H3 ImageNot run. This replaces the person with a made-up character, never a real or named person. Use only your own photo or one from someone who agreed.
Swap me for an invented character
Keep my pose, the park background and the daylight exactly the same. Replace me with an invented original character: a fantasy ranger with braided silver hair, a green hooded cloak and leather armour, in a realistic painted style. The face is a new made-up face. No text, logos or watermarks.
Pricing on this site
Credit packs or a monthly plan.
Credits come in one-time packs ($4.99 for 50 credits, $9.99 for 120 credits, $19.99 for 300 credits); each image costs 10 credits. Or pick a plan from $9.90 a month (500 credits), cancel anytime.
Refund policy- Basic · $9.90/month500 credits a month · about 50 images · $0.20 per image
- Pro · $19.90/month1500 credits a month · about 150 images · $0.13 per image
Billed every month and renews automatically. Cancel anytime in Settings; you keep the plan until the end of the month you paid for. Each month's credits are for that month; unused ones expire.
Refund policyPhoto tools on this site
Edit, restore and restyle your own photo with this site’s image model
- Alternatives and AI toolsOpen
MiniMax H3 ImageOpen
LongCat VideoOpen
Adobe Firefly alternativeOpen
EasyFaceSwap alternativeOpen
Vidwud face swap alternativeOpen
MeiGen AI alternativeOpen
SnapGen AI alternativeOpen
Canva headshot generator alternativeOpen
QuickAIHeadshots alternativeOpen
Betterpic alternativeOpen
Everygen alternativeOpen
PhotoCraft alternativeOpen
Free clipping AIOpen
What is KixyOpen
OpenShorts alternativeOpen
Snazo alternativeOpen
Vikas edit alternativeOpen
Seedream 5.5Open
PromptSeen alternativeOpen
Prompt AI EditzOpen
Ronytica alternativeOpen
Pixlr alternativeOpen
Free AI image generator, no restrictionsOpen
MiniMax H3 Image ComfyUI workflowsOpen
SeedanceOpen
LTX 2.5Open
LongCat Video Avatar 1.6Open
PAN2Open
FastFrames LOpen
Swagga AI websiteOpen
Wan 2.2 and Wan 2.1Open
M4 Editor alternativeOpen
Harish Edits alternativeOpen- AI photo effectsOpen
- All AI photo effectsOpen
FAQ
Questions, answered
What is the MiniMax H3 Image character swap?
A community model on Hugging Face (akhaliq/MiniMax-H3-Image-Character-Swap) that fuses a character-swap LoRA into MiniMax’s H3 Image model. You give it a scene and a character reference, and it is meant to swap the named person while keeping the rest of the scene. The details are the author’s claims.
Is the MiniMax H3 Image character swap official, or made by MiniMax?
No. It is published by the Hugging Face user akhaliq, built on MiniMax’s open H3 Image model and a LoRA by Akatz Labs. We found no MiniMax statement endorsing it, and the author’s profile does not say they work for MiniMax or Hugging Face.
How much VRAM does the character swap LoRA need, and can I run it on my own GPU?
The model card says a full local run with diffusers needs about 140 GB of VRAM, or less if the text encoder runs on another machine. For ComfyUI it offers a 40 GB checkpoint and states no VRAM figure. Most home cards will need a hosted route instead.
Is it free, and can I use it commercially?
The files can be downloaded, but they fall under MiniMax’s H3 Image community licence. It excludes the EU, the UK, South Korea and the United States from its territory, and needs MiniMax’s written permission above 1 million USD of yearly revenue or for hosted services. Read the licence text; this is not legal advice.
Is it safe to download the files and the ComfyUI workflow?
We found no report of a problem and did not open the merge script or workflow file, so we cannot vouch for them. General advice from Hugging Face and Comfy Org: load models only from sources you trust, prefer safetensors files over pickle files, add custom nodes only from sources you trust, and read any script before you run it.
Can I do an AI character swap or replace a person in a photo on this site?
Not a real one. This site’s editor takes one photo and a written request, has no second reference picture, and refuses face swaps and look-alikes of real people. It can swap your outfit, the background or the style, or replace the person with an invented character, in about 37 seconds for 10 credits.
Is it free here?
Partly: new accounts get 10 free credits after signing in, enough for 1 free image (once per account, valid for 7 days, while the day's new-account gifts last). Each image costs 10 credits, from $0.13 per image on a monthly plan.
Can I swap a person in a video?
Not here: this site makes still images only. The underlying Akatz LoRA was built for swapping a person in a source video, and the repo here is its image version; we did not test either.
Want a swap-style edit of your own photo?
Upload it, write the change, and download the result in about 37 seconds.
Edit a photo hereIndependent site. Not affiliated with MiniMax, Akatz Labs, the repo’s author, Hugging Face, Comfy Org or RunComfy, and this site does not run H3 Image or the character swap. Facts come from the pages linked under Sources, read on 2026-11-19; the model’s claims are the author’s and not tested by us. The licence summary is not legal advice. Only use photos of yourself or of people who agreed.