MiniMax releases and open-sources the H3-Image image model.
Start here
MiniMax released and open-sourced the H3-Image image model on November 10. A single checkpoint supports text-to-image and editing with up to 16 reference images, with native output at 4096×4096 (approximately 16MP) and 6-step sampling without CFG. The company claims it has the strongest instruction following among open-source models, runs at 5 times the speed of mainstream models, and is priced at 50%–70% of mainstream competitors' API prices. For the first week, requests at 16MP and above are billed at 50% of the official price. The license excludes the EU, UK, South Korea, and the US; annual revenue over $1 million or providing hosted services to third parties requires written permission. ComfyUI 0.45.0 natively supports it. The pDMD 2-NFE LoRA reduces sampling from 6 steps to 2 steps, cutting 4096×4096 generation on H200 from 25.1 seconds to 17.4 seconds, and editing on RTX 4070 from an average of 38.83 seconds to 27.62 seconds. Q2_K quantization reduces VRAM requirements from 68GB to 6.4GB, allowing 4K local generation on 8GB GPUs. Moore Threads MTT S5000 completed adaptation on the day of open-sourcing. TMTPost's real-world tests report speed ahead of GPT Image 2.6 and Seedream 5.5, but complex instruction execution is unstable. The official claim is 10-second generation; actual tests show standard mode at about 25–30 seconds and high-speed mode at 10–15 seconds.
Written by AI from the reports · updated 4 hours ago
Developments
16 developments
- Nov 19 10:12 · 1 reportMiniMax H3 Image Edit Workflow TutorialStable Diffusion Art 教程:MiniMax H3 Image Edit Workflow Tutorial: Converting Anime Images to Realistic Photos
- Nov 18 21:44 · 1 reportMiniMax H3 Image generates native 16MP images in ComfyUI (with workflow)Stable Diffusion Art 教程:MiniMax H3 Image generates native 16MP images in ComfyUI (with workflow)
- Nov 16 05:15 · 1 reportHands-on test of the pDMD 2-step LoRA for MiniMax-H3-Imageかみもと · 本地AI视频工作流:Hands-on test of the pDMD 2-step ComfyUI workflow for MiniMax-H3-Image: generation time on an RTX 4070 reduced by about 29%
- Nov 16 02:42 · 1 reportMiniMax H3 Image ComfyUI Workflow TutorialAitrepreneur · YouTube 创作视频:MiniMax H3 Image Local ComfyUI Workflow Tutorial: 4K Generation and Multi-Image Editing
- Nov 15 04:46 · 1 reportMiniMax H3 Image quantized version low-VRAM tutorialSuSu_酥酥:MiniMax H3 Image quantized version reduces VRAM from 68GB to 6.4GB; 8GB graphics cards can run 4K image generation.
- Nov 14 09:31 · 1 reportMiniMax Releases and Opens Up H3-Image Weights钛媒体:MiniMax H3-Image Hands-On Comparison Against GPT Image 2.6 and Seedream 5.5: Faster Speed, but Unstable Execution of Complex Instructions
- Nov 12 00:32 · 1 reportMiniMax H3 Image Model Installation and Usage TutorialStable Diffusion Tutorials · 创作教程:Minimax H3 Image: Image Gen & Editing (BF16/FP8/GGUF/INT8/W4A8)
- Nov 11 15:15 · 1 reportRunning MiniMax-H3-Image GGUF evaluation on M1 Maxlilting channel (English) · 创作分享:MiniMax-H3-Image GGUF tested on M1 Max 64GB: 4MP anime illustration generated in about 5 minutes
- Nov 11 09:56 · 1 reportMoore Threads completes MiniMax H3-Image adaptation.IT之家:Day-0 support: Moore Threads completes adaptation for MiniMax H3-Image image generation model.
- Nov 11 09:44 · 1 reportSharing MiniMax H3 Image usage experience and setup tips.Reddit StableDiffusion · 创作者与工作流分享:MiniMax H3 Image first-day hands-on: Author shares Fuobe settings and key parameters.
- Nov 11 09:38 · 1 reportMiniMax-H3-Image trial and ComfyUI workflow sharingnote · comfylab · 图像视频创作分享:ComfyUI hands-on test of MiniMax-H3-Image: directly generates 16-megapixel images
- Nov 11 07:06 · 1 reportMiniMax H3 Image Hands-on Experience: Key Settings and ObservationsReddit comfyui · 创作与工作流分享:MiniMax H3 Image Hands-on Experience: Key Settings and Observations
- Nov 11 06:08 · 1 reportMiniMax-H3-Image license restricts use in the US, UK, South Korea, and the EU.Reddit StableDiffusion · 创作者与工作流分享:MiniMax-H3-Image license restricts use in the US, UK, South Korea, and the EU.
- Nov 11 00:39 · 3 reportsComfyUI natively supports the MiniMax H3 Image model.Reddit comfyui · 创作与工作流分享:MiniMax H3 Image receives day-0 native support in ComfyUI
- Nov 10 08:00 · 3 reportsMiniMax releases H3-Image and open-weights it.量子位 · 搜狐公开专栏:MiniMax open-sources the H3-Image image model, supporting native 4K and up to 16 reference images
- Nov 10 08:00 · 1 reportShare MiniMax-H3-Image 2-step LoRA workflowComfyUI Wiki · 工作流与开源创作分享:MiniMax-H3-Image integrates pDMD 2-NFE LoRA, completing sampling in 2 steps
Report timeline
Follow the reports to see the event from other sides.
- Stable Diffusion Art 教程PicksMiniMax H3 Image Edit Workflow Tutorial: Converting Anime Images to Realistic Photos
Stable Diffusion Art publishes a tutorial showing how to convert anime images into realistic photos in ComfyUI using the MiniMax H3 Image Edit workflow. The workflow uses only native nodes, preserving character pose and composition through the vision-language pathway and reference latent pathway, without needing a ControlNet stack. Optionally enable the Qwen 3.5 prompt enhancer, supporting scene-level custom edits.
- Stable Diffusion Art 教程PicksMiniMax H3 Image generates native 16MP images in ComfyUI (with workflow)
MiniMax H3 Image can natively generate approximately 16.8MP images in ComfyUI, completed in 6 steps, without external super-resolution. The tutorial provides a complete workflow JSON and gives quantized model options: diffusion model 21GB, text encoder 15.7GB, image VAE 1.17GB, and can run on a 24GB GPU.
- かみもと · 本地AI视频工作流PicksHands-on test of the pDMD 2-step ComfyUI workflow for MiniMax-H3-Image: generation time on an RTX 4070 reduced by about 29%
The pDMD 2-step LoRA for MiniMax-H3-Image, using ComfyUI 0.45.0 and an RTX 4070 12GB, reduced the average generation time for 9 image edits from 38.83 seconds to 27.62 seconds, a reduction of about 28.9%.
- Aitrepreneur · YouTube 创作视频MiniMax H3 Image Local ComfyUI Workflow Tutorial: 4K Generation and Multi-Image Editing
MiniMax H3 Image can run locally in ComfyUI, supporting native 4K image generation, editing existing images, and combining up to 16 reference images. After a week of hands-on testing, the author provides one-click installation options for local and RunPod setups, covering text-to-image generation, multi-reference editing, inpainting, outpainting, and a built-in prompt enhancer.
- SuSu_酥酥PicksMiniMax H3 Image quantized version reduces VRAM from 68GB to 6.4GB; 8GB graphics cards can run 4K image generation.
The open-source community has quantized MiniMax H3 Image to Q2_K, reducing VRAM requirements from 68GB to 6.4GB, allowing 8GB graphics cards to run 4K image generation locally. The quantized version natively outputs 4096×4096 and 21:9 widescreen, official distillation produces images in 6 steps, integrates generation and editing, and supports up to 16 reference images.
- 钛媒体PicksMiniMax H3-Image Hands-On Comparison Against GPT Image 2.6 and Seedream 5.5: Faster Speed, but Unstable Execution of Complex Instructions
MiniMax released the H3-Image model weights on November 10. The company says it can generate a 16-megapixel high-resolution image in 10 seconds, at a price about 50%-70% of mainstream tools in the same tier.
- 量子位 · 搜狐公开专栏PicksMiniMax open-sources the H3-Image image model, supporting native 4K and up to 16 reference images
MiniMax has open-sourced the H3-Image image generation model, which supports 16MP (4096×4096) native 4K output and up to 16 reference images as input. In the author's hands-on testing, it can generate complex images such as CAD three-view drawings, mechanical exploded views, and advanced calculus exam papers, with web-based 4K generation taking about 10–20 seconds. API pricing is 50% to 70% of mainstream competitors, and 16MP requests receive an additional 50% discount during the first week.
- Stable Diffusion Tutorials · 创作教程Minimax H3 Image: Image Gen & Editing (BF16/FP8/GGUF/INT8/W4A8)
- lilting channel (English) · 创作分享PicksMiniMax-H3-Image GGUF tested on M1 Max 64GB: 4MP anime illustration generated in about 5 minutes
The author tested the community GGUF quantized version (Q4_K_M, 11.57GB) of MiniMax-H3-Image on M1 Max 64GB, generating a 4MP anime illustration in about 5 minutes, with 6 sampling steps at about 47 seconds per step.
- IT之家Day-0 support: Moore Threads completes adaptation for MiniMax H3-Image image generation model.
Moore Threads' MTT S5000 AI computing card and MUSA software stack completed adaptation for the MiniMax H3-Image image generation model on the day of open-sourcing. H3-Image can generate 4096×4096 pixel (approximately 16MP/4K) images in 10 seconds, with speed reaching 5 times the mainstream level on the market. Moore Threads reused components such as SGLang, MATE, and MCCL to complete the adaptation, laying the foundation for Day-0 adaptation of future image and video generation models.
- Reddit StableDiffusion · 创作者与工作流分享MiniMax H3 Image first-day hands-on: Author shares Fuobe settings and key parameters.
After testing MiniMax H3 Image, the author shares key settings: the same checkpoint can do both text-to-image and editing; editing requires connecting a VAE; reference images are written into the prompt as <Picture 1> etc. in connection order. The model natively supports 4096x4096, and can generate images with 6 steps, CFG 1, and no negative prompt; longer prompts work better; it has received Fuobe Day 0 support and can be tried for free.
- note · comfylab · 图像视频创作分享ComfyUI hands-on test of MiniMax-H3-Image: directly generates 16-megapixel images
MiniMax releases MiniMax-H3-Image, a pure image model based on the H3 video model, supporting direct generation at approximately 16 million pixels and editing with up to 16 reference images. The author tested it on an RTX 4070 Ti SUPER: 4MP generation averages about 9 seconds, 8MP about 17 seconds, and 16MP about 37 seconds. ComfyUI 0.45.0 already includes an official template. The model uses a community license and is available in regions such as Japan, but is excluded from the EU, UK, South Korea, and the US.
- Reddit comfyui · 创作与工作流分享MiniMax H3 Image Hands-on Experience: Key Settings and Observations
The author shares their experience using MiniMax H3 Image on the day of its release: the same checkpoint supports text-to-image and editing; editing requires connecting VAE; prompts should be long, describing lighting, camera, and subject position; native resolution is 4096x4096. Fuobe has provided Day 0 support, source code at https://github.com/workingprototype/Fuobe。.
- Reddit StableDiffusion · 创作者与工作流分享PicksMiniMax-H3-Image license restricts use in the US, UK, South Korea, and the EU.
The MiniMax-H3-Image license explicitly prohibits use in the EU, UK, South Korea, and the US. The author reminds local creators to be aware of compliance risks. The license text lists excluded regions, and the link for requesting an exception currently returns 404.
- comfyorg · YouTube 创作视频ComfyUI natively supports the MiniMax H3 Image image model.
ComfyUI natively supports the MiniMax H3 Image open-weight image model, providing workflow templates for text-to-image and reference image editing. This model supports text-to-image and reference image editing with a single checkpoint, six-step sampling without parameter tuning, native output up to 4096×4096, wide 21:9 generated by size, prompt limit of 4096 tokens, and can run on local hardware or Partner Nodes.
- Reddit comfyui · 创作与工作流分享PicksMiniMax-H3-Image released, ComfyUI already has native workflows.
MiniMax releases MiniMax-H3-Image, and ComfyUI already provides two native workflows. A single checkpoint supports both text generation and image-reference editing, generating approximately 16MP per pass with six-step sampling, without the need for separate upscaling. ComfyUI needs to be updated to 0.45, and the model files include the Qwen3-VL 32B text encoder, image VAE, and Qwen3.5 9B prompt expander.
- Reddit comfyui · 创作与工作流分享PicksMiniMax H3 Image receives day-0 native support in ComfyUI
MiniMax H3 Image receives day-0 native support in ComfyUI, with a single checkpoint supporting both text-to-image and reference image editing. The model is CFG-distilled, requires 6 sampling steps and no negative prompt, natively outputs resolutions such as 4096×4096 (1:1), and supports multi-reference image editing and prompts up to 4096 tokens.
- IT之家MiniMax open-sources H3-Image image generation model, capable of generating a 4K image in 10 seconds.
MiniMax releases and open-weights the H3-Image image generation model, which can generate high-resolution images of 4096×4096 pixels (approximately 16MP) within 10 seconds. The model supports text-to-image, image reference, and image editing. The company claims it has the strongest instruction-following capability among open-source models, supports up to 16 reference images, and is 5 times faster than mainstream models on the market. During the first week of release, generation requests for 16MP and above are offered at 50% off the official price.
- ComfyUI Wiki · 工作流与开源创作分享MiniMax-H3-Image integrates pDMD 2-NFE LoRA, completing sampling in 2 steps
akhaliq merged the pDMD 2-NFE LoRA into the Comfy-Org checkpoint, reducing MiniMax-H3-Image sampling from 6 steps to 2 steps; a single 4096x4096 image on H200 went from 25.1 seconds to 17.4 seconds. The LoRA has been pre-merged into the 40GB weight file, or the 2GB LoRA can be loaded separately into an existing checkpoint; 2 steps work for most prompts, but graphic prompts are recommended at 3 steps.
- ComfyUI Wiki · 工作流与开源创作分享PicksMiniMax releases open-source image model MiniMax-H3-Image, supporting editing with up to 16 reference images.
On 2026-11-10, MiniMax released the open-weight image model MiniMax-H3-Image. A single checkpoint supports both text-to-image and editing with up to 16 reference images, with approximately 16 MP output and 6-step sampling without CFG. The model is based on a 33.1B DiT and the Qwen3-VL-32B text encoder; text generation is a core selling point. On the same day, ComfyUI v0.45.0 added native support, with official T2I and Edit templates built in.
Heat trend for this event
- Current
- 9
- Peak
- 38Nov 12 20:00
- 24h change
- –
The trend compares only sources observed the whole time, so the range may be smaller than the current heat figure. Hover or tap the chart for each hour. Left and right arrow keys move through it.