ComfyUI v0.29.0 Text Generation Bug Causes Gemma to Leak Internal Planning
jjjnnnxxx · reddit · 2026-08-02
Users report that after upgrading to ComfyUI v0.29.0, the built-in TextGenerate node experiences a bug. When using the Gemma model for prompt generation, the model unexpectedly outputs internal planning/reasoning processes alongside the final text.
The issue has been raised on GitHub (#15143) and is confirmed to be a problem with the node itself rather than the Gemma model's intended behavior. Regardless of how system instructions or prompts are configured, the node continues to leak internal planning information.
More from Multimodal
- A Fully Reusable Video Prompt: Turning Four Reference Photos into a 15-Second Surrealist Short — umesh_ai · 2026-10-17
- Runway video packs cinematic destruction into 15 seconds, sparks realism debate — umesh_ai · 2026-10-17
- MOSS Image Sampler Tests: dpmpp_sde + radiance_shift Follows Prompts Best at cfg 3 — mikemend · 2026-10-17
- AI-generated Britney Spears fashion spreads with giant sculptural outfits go viral — OVolosin82152 · 2026-10-17
- Pika partners with OpenAI to bring its full creative studio into ChatGPT — TomLikesRobots · 2026-10-17
- Vercel's Rauch rebuilds Tesla Roadster demo as playable, highly accurate sim — nicolascraske · 2026-10-17