DEV Community

AI-KSK
AI-KSK

Posted on

Boogu 0.1 in ComfyUI: Three Open Models in One Release, Tested Against Qwen-Class Baselines

#comfyui #ai #tutorial

Boogu 0.1 opened three models at once, around the 10B class. This episode is a first-round comparison against Qwen-series baselines, plus three ComfyUI routes: single-image editing, Turbo text-to-image and the Base text-to-image graph.

Evidence note: no workflow JSON was shipped and parameters were not parsed; routes and links come from the episode description and the verified workflow pages.

The three graphs

Single-image editing (https://www.runninghub.ai/post/2067856537388933122/?inviteCode=rh-v1111) — edit an existing image rather than generate from scratch.

Turbo ultra-fast text-to-image (https://www.runninghub.ai/post/2067853157933338626/?inviteCode=rh-v1111) — the speed route; useful for exploring prompts before a heavier render.

Base text-to-image (https://www.runninghub.ai/post/2067856528262127617/?inviteCode=rh-v1111) — the quality route with the base checkpoint.

What the comparison is actually about

The interesting question with a new 10B-class family is not "is it better than Qwen" in the abstract, but where it wins: prompt adherence, text rendering, or edit fidelity. The episode's first-round framing is exactly that — a baseline comparison, not a victory lap.

How to approach the three graphs

  1. Start with Base to see what the model does at default settings.
  2. Switch to Turbo when you are iterating on prompt wording.
  3. Use the editing graph to test whether it preserves identity while changing context.

Limits

  • A first-round comparison is not a benchmark; treat the ranking as directional.
  • 10B-class size implies real VRAM needs.
  • No JSON/parameters parsed for this episode.
  • Three models in one release does not mean three equally strong models.

Episode resources

Top comments (0)