{"slug": "minimax-h3-image-native-4k-at-16-8mp-in-six-steps-three-comfyui-workflows", "title": "MiniMax H3-Image: Native 4K at 16.8MP in Six Steps - Three ComfyUI Workflows Compared", "summary": "A developer published three ComfyUI workflows for MiniMax's H3-Image model, which generates natively at 4K-class resolution (about 16.8 megapixels) in six sampling steps rather than upscaling. The workflows cover text-to-image (six nodes), single-image editing (48 nodes), and a 16-image reference edit (48 nodes) for character and product consistency, and ship as JSON graphs without bundled model weights. All three were tested as online workflows on RunningHub, with the open question being whether the 16-image branch preserves identity when references conflict.", "body_md": "MiniMax H3-Image: native 4K at 16.8 megapixels in six sampling steps (three ComfyUI workflows)\n\nMiniMax's H3-Image is unusual for one reason: it does not upscale its way to a big picture. It generates natively at 4K-class resolution - about 16.8 megapixels - and it does it in six sampling steps. This episode tests that claim with three separate ComfyUI workflows, each built for a different job.\n\nWho this is for\n\nThe three workflows, and when to use each\n\nText to image - `AIKSK_H3_Image_文生图.json`\n\nThe simplest chain: prompt in, 6 steps of sampling, 4K image out. Six nodes total, so you can read the whole graph in a minute. Use it when the image does not exist yet.\n\nOnline: [https://www.runninghub.ai/zh-cn/post/2120808839725920258/?inviteCode=rh-v1111](https://www.runninghub.ai/zh-cn/post/2120808839725920258/?inviteCode=rh-v1111)\n\nSingle image edit - `H3_Image_单图像编辑.json`\n\n48 nodes. One reference image goes in with an instruction, and the model re-renders while keeping the parts you did not ask it to change. This is the workflow for \"same photo, different weather / different light / fix this corner\".\n\nOnline: [https://www.runninghub.ai/zh-cn/post/2120808831056293890/?inviteCode=rh-v1111](https://www.runninghub.ai/zh-cn/post/2120808831056293890/?inviteCode=rh-v1111)\n\n16-image reference edit - `H3_Image_16图编辑.json`\n\nAlso 48 nodes, but the reference side is a bank of up to sixteen images. That is what you want for character consistency across angles, or for feeding a whole product set into one edit so the result matches all of them.\n\nOnline: [https://www.runninghub.ai/zh-cn/post/2120808834377023490/?inviteCode=rh-v1111](https://www.runninghub.ai/zh-cn/post/2120808834377023490/?inviteCode=rh-v1111)\n\nWhat is inside the JSON\n\n`qwen3.8_27b_w4a8` for the prompt path.`minimax_h3_image_pruned_int8_convrot`, `minimax_h3_image_vae_fp16`, `qwen3.5_9b_w4a8_gptq`, `qwen3vl_32b_minimax_h3_int8_convrot`.\nNo model weights are bundled in the repos. You bring the weights; the workflow carries the wiring, the sampling settings and the conditioning structure.\n\nKnown limits\n\nHow to try it without a local install\n\nAll three run as online workflows on RunningHub, which is what we used for the tests in the episode - same graph, no GPU required on your side:\n\nEpisode links\n\nWhere this goes next\n\nThe interesting question is not whether 4K native generation looks good in a hero shot - it does. It is whether the 16-image branch holds identity when the references disagree with each other. That is the test worth running yourself.", "url": "https://wpnews.pro/news/minimax-h3-image-native-4k-at-16-8mp-in-six-steps-three-comfyui-workflows", "canonical_source": "https://dev.to/ai-ksk/minimax-h3-image-native-4k-at-168mp-in-six-steps-three-comfyui-workflows-compared-2am2", "published_at": "2026-11-21 10:09:34+00:00", "updated_at": "2026-11-21 10:11:54.086770+00:00", "lang": "en", "topics": ["generative-ai", "ai-tools", "computer-vision", "ai-products"], "entities": ["MiniMax", "H3-Image", "ComfyUI", "RunningHub", "Qwen"], "also_reported_by": [], "takeaway": "MiniMax H3-Image generates natively at roughly 16.8 megapixels in six sampling steps, and the shared ComfyUI graphs pair it with int8-pruned weights plus Qwen text and vision encoders, leaving weights to the user.", "recommendation": {"advice": "Run the six-node text-to-image graph first to validate H3-Image at 4K on your hardware, then move to the 48-node edit graphs and test the 16-image branch for identity drift when references disagree.", "playbook": "inference-serving", "playbook_url": "https://wpnews.pro/playbooks/inference-serving"}, "alternates": {"html": "https://wpnews.pro/news/minimax-h3-image-native-4k-at-16-8mp-in-six-steps-three-comfyui-workflows", "markdown": "https://wpnews.pro/news/minimax-h3-image-native-4k-at-16-8mp-in-six-steps-three-comfyui-workflows.md", "text": "https://wpnews.pro/news/minimax-h3-image-native-4k-at-16-8mp-in-six-steps-three-comfyui-workflows.txt", "jsonld": "https://wpnews.pro/news/minimax-h3-image-native-4k-at-16-8mp-in-six-steps-three-comfyui-workflows.jsonld"}}