Ideogram V4
V4 LoRA trainer teaching subjects, characters, and styles from captioned zip archives.
Studio is in early access; availability resolves per project after sign-in.
- Endpoints
- 9
- Eligible
- 4
- Input schemas imported
- 9 / 9
- Delivery
- fal.ai
- Weights
- Not stated
- Developer
- Not stated in source
Overview
The Ideogram V4 trainer fine-tunes compact LoRA adapters that teach the base model new subjects, characters, objects, or visual styles without full retraining. Training runs on captioned images from a single zip archive with trigger phrases; weights drop into the fal V4 endpoint or ComfyUI. PNG, JPG, JPEG, and WebP process; HEIC, HEIF, and AVIF are silently skipped. Base, fast, image-to-image, instant, tiling, and LoRA endpoints join the trainer in the 9-member family.
Capabilities
- Compact LoRA adapters
- Caption-driven training
- Dual deploy targets
- Auto resolution selection
Best for
- Subject teaching
- Character consistency
- Style adaptation
Use cases
- Zip-archive training runs
- Trigger-phrase styles
- ComfyUI weight drops
Endpoints
5 under review · 4 eligible. Cataloged is not the same as executable: Studio resolves which endpoints a project can run.
| Endpoint | Task | Catalog status | Inputs |
|---|---|---|---|
ideogram/v4 | unknown | Under review | 10 fields · requires prompt |
ideogram/v4/fast | unknown | Under review | 9 fields · requires prompt |
ideogram/v4/image-to-image | image to image | Eligible | 12 fields · requires prompt, image_url |
ideogram/v4/image-to-image/lora | image to image | Eligible | 13 fields · requires prompt, image_url |
ideogram/v4/instant | unknown | Under review | 8 fields · requires prompt |
ideogram/v4/lora | lora training | Eligible | 11 fields · requires prompt |
ideogram/v4/tiling | unknown | Under review | 14 fields · requires prompt |
ideogram/v4/tiling/lora | lora training | Eligible | 15 fields · requires prompt |
ideogram/v4/trainer | unknown | Under review | 6 fields · requires images_data_url |
Questions
Full retraining?
No: compact added weights only.
Which formats?
PNG, JPG, JPEG, and WebP; others skipped.