MiniMax H3 Image Evening Report 2026-11-20 | First Recommendation: Simple t2i & i2i with Preview — H3 Image Workflow

This issue reviews 2 new versions of H3 Image, including 1 S-rated and 1 C-rated items. The top recommendation is v1.0pro of Simple t2i & i2i with Preview: in addition to the existing text-to-image generation, real-time preview, and prompt enhancement features, it adds image reference and editing capabilities using up to 3 reference images. All required models, nodes, and sampling parameters are fully documented. The H3 Image version of the Skeletor character LoRA lacks specific documentation and is rated C.

Zhenzhen's AI Radar · MiniMax H3 Image · 2026-11-20 Evening Report

This issue features 2 new versions: Strongly Recommended 1 · Recommended 0 · Worth Watching 0 · Listed Only 1

Editor's Note This issue reviews 2 new versions of H3 Image, including 1 S-rated and 1 C-rated items. The top recommendation is v1.0pro of Simple t2i & i2i with Preview: in addition to the existing text-to-image generation, real-time preview, and prompt enhancement features, it adds image reference and editing capabilities using up to 3 reference images. All required models, nodes, and sampling parameters are fully documented. The H3 Image version of the Skeletor character LoRA lacks specific documentation and is rated C.

Strongly Recommended (1)

[S] Simple t2i & i2i with Preview — H3 Image Workflow · v1.0pro

Workflow · Version Update · Author spacejumper2021824 · Published 11-20 12:17 · Downloads 46

A concise ComfyUI workflow for MiniMax H3 Image. The v1.0pro version adds image-to-image capabilities on top of text-to-image, real-time preview, and prompt enhancement: it supports up to 3 reference images for image referencing and editing.

Images

Example 1809x1209

Reasons for Recommendation

  • Version Update: v1.0 offers text-to-image with preview; v1.0plus adds prompt enhancement and preset sampling combinations; v1.0pro further adds Image Reference & Editing, supporting up to 3 reference images by default. More references require manual workflow modification.
  • Real-time Preview: Uses KJNodes' ModelPreviewOverrideKJ with the taeh3_image small VAE (4.96 MB) to display visuals during generation, eliminating the need to wait for completion.
  • Actionable Instructions: The card lists required model filenames, two custom node packages, directory locations for each file, and two sets of sampling parameters. It is free, and the license permits commercial use and sales.

How to Use

  • Install ComfyUI-KJNodes and rgthree-comfy via ComfyUI Manager; place models minimax_h3_image_pruned_int8_convrot, qwen3vl_32b_minimax_h3_int8_convrot (text encoder), and minimax_h3_image_vae_fp16, along with taeh3_image.safetensors, into models/vae_approx/.
  • Official recommendation: euler / simple / 6 steps; the author suggests switching to sa_solver / sgm_uniform / 5 steps if realistic portrait skin shows excessive noise or smearing.
  • Default resolution is 4096×4096 (16 MP, 1:1), adjustable via the ResolutionSelector node; for the pro version, note the toggle between "Custom Resolution" and "Follow Reference Image Resolution."
  • Prompt enhancement can be manually disabled; it uses qwen3.5_9b_w4a8_gptq, though the author notes it can also be replaced with MOSS Image's text encoder.

Risks and Limitations

  • Only 2 example images are provided, with no before-and-after comparisons showing reference image editing.
  • The default 16 MP output has high VRAM requirements; resolution should be lowered according to your hardware. The archive contains workflows; verify contents before installation.

Listed Only (1)

  • Skeletor - Masters of the Universe · v1.0 - H3 Image — LoRA: A character LoRA for Skeletor, the villain from "He-Man and the Masters of the Universe," in the H3 Image version; the same page also lists versions for MOSS, Flux.3 Klein, Qwen 2.1, LTXV, Krea 2, etc. Example 5472x3072

Data: Civitai Public API (Base Model MiniMax H3), inclusion window 11-20 08:31 → 11-20 21:01 (Beijing Time). Recommendations have been reviewed by editors; download counts reflect attention levels only.

Civitai

Explore the latest AI Radar updates