Home Creators Posts Import Register Favorites Logout

Downloads

Content

Video : https://youtu.be/1gNpW57EE2k

Hey everyone! This is good model personal I tested and gen many images this 2 days with this model.

New video is up (https://youtu.be/1gNpW57EE2k) , and this one is for anyone who has been waiting for an open-weights image model that doesn't need a node pack and a prayer to run. MiniMax H3 Image dropped with the claim of being state of the art for open models, so I ran the whole thing locally and stress-tested it instead of just showing pretty pictures.

The short answer: it's genuinely strong. Native nodes in ComfyUI, six to eight steps, CFG at one, up to 4096 by 4096 straight out. My favorite moment was the mirror physics test, a guy fixing his tie in front of a bathroom mirror, and the reflection actually follows mirror rules. Elbows, hands, the lot. Text rendering on a screen-printed poster held up too, line by line, spelling and grain and everything.

Two free workflows are attached here, the text-to-image template and the image-edit template, both straight from what's in the video. Use them as-is. The full list of JSON prompt examples from every test is in the video description and the pinned comment.

For Patreon Supporters , new additional post
https://www.patreon.com/aifuturetech/posts/175034979

Model links, if you want the source repos:

- https://huggingface.co/MiniMaxAI/MiniMax-H3-Image

- https://huggingface.co/MiniMaxAI/MiniMax-H3-Image-PE/tree/main/prompt

- https://docs.comfy.org/tutorials/image/minimax/minimax-h3-image

- https://huggingface.co/Comfy-Org/MiniMax-H3-Image/

Thanks for watching, and see you in the next one.

— Benji

Files

Minimax H3 Image Release SOTA Open Weights Model! You Can Run Locally!

MiniMax H3 Image just dropped as open weights, and in this video I run it locally in ComfyUI to see if the state-of-the-art claim holds up. Full setup from the three model files, then real stress tests instead of pretty pictures. Everything runs on native ComfyUI nodes, no custom node packs. Six to eight sampling steps, CFG 1.0, and a distilled model that outputs at up to 4K, 4096 by 4096. I show the INT8 quantized path on a consumer-grade setup and explain which files you can reuse if you already ran the MiniMax H3 video model. The tests push the model on purpose: exact fruit placement across five colored bowls to grade prompt adherence, a screen-printed poster to check text rendering line by line, a bilingual Chinese and English Hong Kong menu board for multilingual accuracy, a logic puzzle poster, and a mirror-reflection physics test. Then the editing side: adding a person into an existing room photo while keeping every cable and pillow in place, a two-reference edit that pulls a character from a character sheet into the same room, and multi-reference cinematic action scenes generated from a single frame's worth of inference. Workflows, model links, and the full JSON prompt examples are below and at the pinned comment. Resources: - MiniMax H3 Image model: https://huggingface.co/MiniMaxAI/MiniMax-H3-Image - MiniMax H3 Image PE (prompt enhancement): https://huggingface.co/MiniMaxAI/MiniMax-H3-Image-PE/tree/main/prompt - ComfyUI official docs and templates: https://docs.comfy.org/tutorials/image/minimax/minimax-h3-image - Comfy-Org quantized releases (BF16, INT8, W4A8): https://huggingface.co/Comfy-Org/MiniMax-H3-Image/ - Free workflow files examples: https://www.patreon.com/aifuturetech/posts/175033165?utm_source=youtube&utm_medium=video&utm_campaign=202611 Calliope - Local-first Agentic Idea-to-video Studio https://github.com/benjiyaya/Calliope Chapters 00:00 - Introduction & Minimax H3 Overview 01:40 - Prompt Enhancement (PE) Model & Agent Setup 04:03 - ComfyUI Workflow & Required Model Files 05:16 - Resolution & Aspect Ratio Settings 05:45 - Sampler Settings & CFG Configuration 07:26 - Test 1: Complex Prompt Adherence (Fruit Bowls) 08:44 - Test 2: Text Rendering & Poster Grain 09:30 - Test 3: Multilingual Text (Chinese & English) 09:59 - Test 4: Logic Puzzle & Mirror Physics 11:07 - Image Editing & Multi-Reference Overview 12:36 - Structuring the JSON Edit Prompt 13:58 - Editing Test: Adding Character to Scene 14:52 - Multi-Reference Test: Character Sheets & Lighting 16:04 - Upscaling & Two-Stage Sampling Workflow 16:53 - Action Shots & Dynamic Poses Test 17:45 - Conclusion & Final Thoughts If You Like tutorial like this, You Can Support Our Work In Patreon: https://www.patreon.com/c/aifuturetech #MiniMaxH3Image #ComfyUI #OpenWeights #imageediting

Comments

No comments found for this post.