Benchmark image models in ComfyUI against a fixed prompt set, keep every result in a permanent gallery, and compare them side by side. Bring your own workflows: each model uses its own ComfyUI workflow per mode (txt2img, img2img, inpaint, …). Your exported files stay untouched; ComfyBench fills in prompt, seed, size and parameters at runtime. Systematic variations: prompt variables, parameter sweeps (cfg, sampler, checkpoints, …) and LoRA combinations. Browser UI: set up models and prompts, start and monitor runs, compare, rate and annotate results. Nothing gets lost: results are versioned, never overwritten, and already generated images are skipped.