Descriptions:
Nerdy Rodent puts the Capybara unified visual creation model through its paces inside ComfyUI, covering the full span of its capabilities: text-to-image generation, image-to-image transformation, dedicated image editing, and video generation — all from a single model that runs on a home PC. The video works through multiple test prompts with default settings of 30 steps at CFG 4, comparing different generation pipelines and upscaling methods including standard image upscale, latent upscale, Ultimate Upscale at 1600×600, and SeedVR2.
A standout finding is Capybara’s unusually low effective denoising threshold for image-to-image work — the model produces meaningful transformations at denoise values as low as 0.2, compared to the 0.5–0.6 typically required by other models. The video also demonstrates the difference between standard image-to-image (which tends to shift the entire composition) and the model’s dedicated image editing mode using a clip vision encode approach, which preserves the surrounding scene while changing only the target subject. Tests at 1920×1080 and video generation at 1280×720 with 121 frames at 24fps round out the coverage.
For anyone exploring unified generation models in ComfyUI as alternatives to running separate specialized tools, this walkthrough provides a practical, settings-focused look at what Capybara can and cannot do — with honest assessments of cases where results fall short of the given prompts.
📺 Source: Nerdy Rodent · Published February 26, 2026
🏷️ Format: Tutorial Demo
