Hammershøi Ideogram 4 LoRA Demo

Compare Ideogram 4 without the LoRA against the Hammershoi-Ideogram4-LoRA at two training checkpoints, step 2000 and step 2500, using the same prompt and seed.

Describe a subject, pick the framing and lighting, and optionally place objects with bounding boxes. The Space builds the structured JSON caption Ideogram 4 expects, with the hmrsh trigger for the LoRA cells and without it for the base cell.

Sign in with Hugging Face once; generations run on your Inference Providers quota. Uncheck cells you don't want to spend quota on.

Images are AI-generated. Ideogram 4 and this LoRA are for non-commercial use only under the Ideogram Non-Commercial Model Agreement. This is a community project, not an official Ideogram product.
Framing
Lighting
Rendering speed
Resolution

Describe an object to place it, then move the sliders. Positions run from 0 to 1000 across the image. A box with no description is ignored.
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
0 1000
Example prompts (click to populate, then press Generate)
Subject description Background Framing Lighting Resolution Box 1 description Top Left Bottom Right Box 2 description Top Left Bottom Right Box 3 description Top Left Bottom Right Box 4 description Top Left Bottom Right

Each image is rendered by fal, which sets its own step count and guidance from the rendering speed, so results won't exactly match the images on the model card.