Ideogram 4 Instant NSFW is a fine-tuned text-to-image checkpoint built on top of ideogram-ai/ideogram-4-fp8 (via the fal/ideogram-v4-fast instant/fast release). [Describe what this finetune changes — e.g. the dataset it was trained on, the style/domain it targets, what makes it different from the base Instant checkpoint.]

Key features

⚡ [Number]-step inference — [describe schedule, e.g. optimized for the Instant/Fast pipeline].
🎯 [Describe any CFG/architecture behavior inherited from base model or changed by finetune]
🧠 [Describe finetuning method] — e.g. LoRA, full finetune, QLoRA, dataset size, training steps.
🧩 Standard Diffusers components — no repository Python code and no trust_remote_code (update this if your finetune requires custom code).
📦 [Describe release scope] — e.g. transformer-only, LoRA weights only, full pipeline.

Usage

This model expects Ideogram 4's structured JSON caption format. Local Diffusers inference does not auto-expand natural-language prompts — provide a complete structured caption, for example:

python

import json
import torch
from diffusers import Ideogram4Pipeline, Ideogram4Transformer2DModel

repo_id = "your-username/ideogram-4-instant-finetune"
components_repo_id = "ideogram-ai/ideogram-4-nf4-diffusers"  # or your preferred shared components repo

transformer = Ideogram4Transformer2DModel.from_pretrained(
    repo_id,
    subfolder="transformer",
    torch_dtype=torch.bfloat16,
)
pipe = Ideogram4Pipeline.from_pretrained(
    components_repo_id,
    transformer=transformer,
    torch_dtype=torch.bfloat16,
)
pipe.to("cuda")

prompt = json.dumps(
    {
        "high_level_description": "Describe your image here.",
        "compositional_deconstruction": {
            "background": "Describe the background.",
            "elements": [
                {"type": "text", "text": "YOUR TEXT", "desc": "Description of the element."}
            ],
        },
    },
    ensure_ascii=False,
    separators=(",", ":"),
)

generator = torch.Generator(device="cuda").manual_seed(42)
image = pipe(
    prompt,
    height=1024,
    width=1024,
    num_inference_steps=20,
    guidance_scale=1.0,
    generator=generator,
).images[0]
image.save("output.png")

Note: Update this snippet to match your finetune's actual inference requirements (precision, number of steps, guidance settings, any custom pipeline code, LoRA loading, etc.).

Training details

Base model: ideogram-ai/ideogram-4-fp8 / fal/ideogram-v4-fast
Method: [LoRA / DreamBooth / full finetune / QAD / etc.]
Dataset: [describe dataset, size, licensing of training data]
Steps / epochs: [fill in]
Hardware: [fill in]
Hyperparameters: [learning rate, batch size, rank if LoRA, etc.]

Repository layout

text

.
├── README.md
├── LICENSE.md
├── NOTICE
└── transformer/  (or lora/ depending on release type)
    ├── config.json
    └── diffusion_pytorch_model.safetensors

Weights and provenance

This model is a fine-tune derived from Ideogram 4 (via ideogram-ai/ideogram-4-fp8 and/or fal/ideogram-v4-fast). [Describe any conversion, quantization, or format changes made during training/export.]

Ideogram 4 was created by Ideogram AI. The Instant/Fast base checkpoint was developed and released by fal. This derivative fine-tune was developed and released by [Your Name/Org] and is not an official Ideogram or fal product, and is not endorsed by Ideogram AI or fal.

License

As a derivative of Ideogram 4, this model inherits the Ideogram 4 Non-Commercial Model Agreement. The complete inherited license is included in LICENSE.md and governs use and redistribution of this model. By downloading or using these weights, you acknowledge and agree to the terms of that license, including its non-commercial use restriction.

Winnougan/Ideogram_Instant_NSFW

Ideogram 4 Instant — NSFW