← All models
Transform your 3D video render into realistic using first frame with Ltx 2.3
Example
Details
- Accepts
- video
Pricing
- Price
- 0.301 cr / megapixel
Prices in credits (1 credit = $0.01).
Data schema
Input
| Field | Type | Description |
|---|---|---|
| seed | — | Random seed for reproducibility. If None, a random seed is chosen. |
| prompt | string | The text prompt, pre-filled with the `3DREAL` trigger the adapter activates on. Describe the photorealistic result after it. The `3DREAL` trigger is always kept, even if you clear it or enable prompt expansion. |
| image_url | — | Optional reference image used as the FIRST frame (image-to-video anchor), matching how the adapter was trained. It defines the photoreal look of frame 0; leave empty to let the render alone drive the result. |
| intensity | stringenum: light, strong, strong-v2 | How hard to push the render toward photorealism. 'light' keeps more of the source render's look and motion; 'strong' transforms more aggressively for a more photoreal result; 'strong-v2' is the newest improved adapter (better multi-scene coherence, audio, talking characters, and higher overall quali… |
| sync_mode | boolean | If True, the media is returned as a data URI inline in the response. Useful for short-lived requests and tests. |
| video_urlrequired | string | The URL of the 3D / CG / game render to make photorealistic. |
| num_frames | integer | The number of output frames. Output duration is num_frames / frames_per_second. The IC-LoRA control budget caps the resolution x frames volume: roughly 6s at 720p or 15s at 480p; a request above it returns a 422 (lower num_frames, or use 480p). |
| resolution | stringenum: 480p, 720p | Output resolution (short side), preserving the source aspect ratio. The IC-LoRA control path caps the resolution x frames volume, so resolution trades against duration: 720p allows up to ~6s (sharper), 480p up to ~15s (softer but longer). A request above the budget is rejected with a clear error — p… |
| video_quality | stringenum: low, medium, high, maximum | The quality preset of the generated video. |
| generate_audio | boolean | Whether to include audio in the returned video. When disabled, the final MP4 is returned without an audio track. |
| guidance_scale | number | Classifier-free guidance scale. The default is tuned for fast, high-quality generation. |
| negative_prompt | string | The negative prompt to steer generation away from. |
| video_write_mode | stringenum: fast, balanced, small | The write mode of the generated video. |
| frames_per_second | number | Frames per second of the generated video. |
| num_inference_steps | integer | Number of inference steps. Defaults to 15 and can be increased up to 30. |
| enable_detail_refine | boolean | Add a second refinement stage for a sharper, higher-resolution result. After the base pass the video latent is 2x latent-upscaled (spatial upscaler) and refined by a short detailer pass on the distilled model. NOTE: this doubles the output width AND height (4x the pixels), so the request costs rough… |
| enable_safety_checker | boolean | Whether to enable the safety checker. |
| detail_refine_strength | number | Denoise strength of the detailer pass, used only when enable_detail_refine is on. Higher regenerates more detail (and can drift further from the base); lower is a lighter touch that stays closer to the base. Default 0.75. |
| enable_prompt_expansion | boolean | When enabled, an AI model expands your prompt into a richer description before generation (off by default sends it as written). The `3DREAL` trigger is always kept, and the final prompt used is returned in the output. |
Output
| Field | Type | Description |
|---|---|---|
| seed | integer | The seed actually used for generation. |
| video | — | The generated video. |
| prompt | string | The prompt used for generation (after any expansion). |