lora-test-grid

LoRA Test Grid

A ComfyUI custom node for comparing LoRAs side by side. Give it a model, CLIP, VAE, a shared prompt, and up to 6 LoRAs — it samples each one with the same seed/settings and tiles the results into a single labeled comparison grid.

Node: LoRA Test Grid

Inputs

• model, clip, vae — standard ComfyUI loaders
• trigger — optional trigger word(s), prepended to the prompt before encoding (e.g. zxq_marilyn)
• prompt — shared test prompt (no negative prompt widget; negative conditioning is generated via zero-out, suited to turbo/distilled models)
• aspect_ratio — dropdown of common presets
• seed, steps, cfg, sampler_name, scheduler — shared sampling settings applied to every tile
• use_master_strength / master_strength — when on, overrides every LoRA slot's individual strength with one shared value
• lora_1 … lora_6 / strength_1 … strength_6 — up to 6 LoRA slots, each with its own strength (ignored if master strength is on)

Output

• grid (IMAGE) — a baseline (no-LoRA) tile plus one tile per active LoRA slot, labeled with filename and strength, tiled 3 columns wide.

Example workflow

workflows/LoRA Test Grid.json — a working setup for a Z-Image-Turbo-style pipeline (UNETLoader/CLIPLoader/VAELoader → LoRA Test Grid → Preview Image), plus a Load Image node for dropping in a dataset/reference photo to compare against the grid output.

Notes

• Reuses ComfyUI's own core node classes internally (LoraLoader, CLIPTextEncode, common_ksampler, VAEDecode), so it should stay compatible as ComfyUI updates.
• The empty latent is built by VAE-encoding a mid-gray image rather than assuming a fixed latent channel count, so it isn't tied to a specific model architecture.
• web/js/lora_test_grid.js hides the per-LoRA strength widgets when use_master_strength is on and restores them when it's off, without disturbing the node's manually-set width.