--- license: other license_name: ideogram-community license_link: https://huggingface.co/ideogram-ai/ideogram-4-fp8/blob/main/LICENSE base_model: - ideogram-ai/ideogram-4-fp8 tags: - ideogram - ideogram-4 - lora - lokr - text-to-image - style - urasawa library_name: diffusers pipeline_tag: text-to-image --- # Naoki Urasawa Ideogram4 v1 LoRA (lokr) adapter trained on **Ideogram 4** (`ideogram-ai/ideogram-4-fp8`) for a **Naoki Urasawa style**. ## Prompt format Ideogram 4 does **not** take a plain-text prompt. It expects a structured **JSON** prompt. This LoRA is activated by setting the style fields: - `style_description.art_style`: `Naoki Urasawa style` - `style_description.medium`: `manga_page` Fill the rest of the JSON schema (`high_level_description`, `style_description.aesthetics/lighting/color_palette`, `compositional_deconstruction`) to describe the scene. ### Example JSON prompt ![sample 3](./sample_03.webp) ```json { "high_level_description": "A humanoid robot stands on a rooftop overlooking a city skyline at night.", "style_description": { "aesthetics": "black_and_white", "lighting": "dramatic, high contrast with strong shadows", "medium": "manga", "art_style": "Naoki Urasawa style", "color_palette": ["#F9F9F9", "#0A0A0A", "#D1D1D1", "#5A5A5A", "#8D8D8D", "#2F2F2F"] }, "compositional_deconstruction": { "background": "A sprawling city skyline with numerous skyscrapers, some lit up, under a starry night sky. The robot stands on a rooftop with a metal railing and scattered industrial equipment.", "elements": [ { "type": "obj", "bbox": [134, 252, 943, 698], "desc": "A tall, humanoid robot with a smooth, featureless head, a segmented torso with a circular chest plate, and articulated limbs. It has a metallic, armored appearance with visible joints and panels. The robot stands upright with a neutral posture.", "color_palette": ["#FAFAFA", "#D3D3D3", "#131313", "#4B4B4B", "#868686"] }, { "type": "text", "bbox": [34, 644, 276, 981], "text": "MAYBE WE SHOULLD HAVE KEPT SOME HUMANS AROUND..", "desc": "A white, oval-shaped speech bubble positioned in the upper right corner. It contains text in a stylized, slightly uneven font, typical of comic book dialogue.", "color_palette": ["#FDFDFD", "#080808", "#363636", "#5F5F5F", "#A8A8A8"] } ] } } ``` ## Specifications | | | |---|---| | Base model | `ideogram-ai/ideogram-4-fp8` | | Architecture | ideogram4 | | Network | lokr (linear 32, alpha 32, factor 8, full rank) | | Training steps | 6000 | | Resolution | 512 / 768 / 1024 | | Optimizer | automagic3 | | Learning rate | 1e-5 | | dtype | bf16 | Keep `art_style` = `Naoki Urasawa style` and `medium` = `manga_page` in every prompt to trigger the LoRA. ## Samples ![sample 1](./sample_01.webp) ![sample 2](./sample_02.webp) ![sample 4](./sample_04.webp) ![sample 5](./sample_05.webp)