DEV Community

AI-KSK
AI-KSK

Posted on

H3 Community Model Updates in ComfyUI: PinkCherry Hybrid 1.8, Eros Beta5/Beta6 and Master Frog V4 in One Mirror

#comfyui #ai #tutorial #videogeneration

H3 Community Model Updates in ComfyUI: PinkCherry Hybrid 1.8, Eros Beta5/Beta6 and Master Frog V4 in One Mirror

This episode is a mirror release, not a JSON pack: it bundles three community-modified MiniMax H3 models and six ComfyUI reference-to-video workflows into one deployable image. The notes below come from the episode subtitles and the published mirror page; no workflow JSON was shipped or parsed for this release, so no node-level parameters are claimed here.

The three model updates

PinkCherry Hybrid 1.8 is a character-focused fine-tune merged with Hybrid. The author's changes target character motion and local detail; in 1.8 the model is no longer limited to its specialty and can also be tested on regular fantasy character motion and reference consistency.

Eros (Beta5 / Beta6) is built by weight grafting: learned features from LTX, Wan and Krea models are introduced into H3. The mirror ships both Beta5 and Beta6 so you can switch; community feedback says Beta6 is stronger on multi-reference inputs, while Beta5 can be more stable depending on the test.

Master Frog V4 is a hybrid-task model that folds Turbo acceleration into the merged weights and is tagged as a specialty model on Civitai.

The six workflows

The mirror contains six ComfyUI graphs that fall into two categories:

  1. Regular generation — the standard reference-to-video route, optionally with a second sampling pass.
  2. Director's console — a node setup that covers most day-to-day directing patterns.

So it looks like six workflows, but functionally it is two families. The quick-start graph lets you pick among the three models (the sample selects models 1, 3 and 5), upload a reference image, and use a mixed node for first-frame control: keep it false to disable, switch to true to enable. Multi-reference images load in the multi-reference slot, and the prompt describes the scene.

Deploying the mirror

  1. Open the mirror page: https://www.compshare.cn/images/djAzmrBXp7ec?referral_code=A7fTQUYY9lCDFJwA6MMvoq
  2. Deploy an instance and pick an available GPU model.
  3. Wait for the image to come up, then open ComfyUI.
  4. Start from the quick-start workflow and switch models or reference slots as needed.

Links

Limits

  • This article describes the mirror contents and the two workflow families; it does not claim exact sampler values, step counts or node names, because no JSON was provided for this episode.
  • Model behaviour depends on prompt, reference images and GPU; the mirror is a deployment convenience, not a quality guarantee.
  • Third-party community models keep their own licenses and usage terms.

Top comments (0)