# Do0p AI Models Directory (14 Foundation Engines)

> Foundational generative AI models powering Do0p Studio pipelines, accessible directly or through specialized tool orchestrations.

Base URL: https://do0p.com  
Tools Catalog: https://do0p.com/tools.md  
API Documentation: https://do0p.com/docs.md  

---

## Foundation Models Summary Table

| Model Name | Slug | Modality | Provider | Cost | Avg Latency |
| :--- | :--- | :---: | :--- | :---: | :---: |
| **Nano Banana 2.1 (Google Flow)** | `nano-banana-2-1` | `IMAGE` | flow-agent | 1 cr | ~0s |
| **Nano Banana Pro** | `nano-banana-pro` | `IMAGE` | flow-agent | 1.2 cr | ~0s |
| **Qwen-Image-2.1-Turbo** | `qwen-image-2-1-turbo` | `IMAGE` | huggingface | 1 cr | ~5s |
| **FLUX.1 Schnell 12B** | `flux-1-schnell` | `IMAGE` | huggingface | 1 cr | ~4s |
| **Google Veo 2 Cinematic** | `google-veo-2` | `VIDEO` | flow-agent | 10 cr | ~22s |
| **Gemini-Omni (Lossless Watermark Inverter)** | `gemini-omni` | `ENHANCEMENT` | flow-agent | 0.2 cr | ~0s |
| **Flux 1.1 Pro (Black Forest Labs)** | `flux-1-1-pro` | `IMAGE` | external-sota | 1.5 cr | ~5s |
| **WAN 2.1 Video & Motion** | `wan-2-1` | `VIDEO` | self-hosted-gpu | 8 cr | ~28s |
| **FaceFusion 3.0 Pro** | `facefusion-3` | `ENHANCEMENT` | self-hosted-gpu | 1.5 cr | ~3s |
| **BRIA RMBG 2.0 (Background Remover)** | `bria-rmbg-2-0` | `ENHANCEMENT` | huggingface | 0.4 cr | ~1s |
| **BiRefNet Studio Matting** | `birefnet-rmbg` | `ENHANCEMENT` | self-hosted-gpu | 0.3 cr | ~1s |
| **CodeFormer 4K Face Restorer** | `codeformer-restore` | `ENHANCEMENT` | huggingface | 0.5 cr | ~2s |
| **Depth Anything V2** | `depth-anything-v2` | `ENHANCEMENT` | huggingface | 0.3 cr | ~1s |
| **Idol 3D Gaussians (.PLY)** | `idol-3d-gaussian` | `3D` | huggingface | 2 cr | ~9s |

---

## Model Details

### Nano Banana 2.1 (Google Flow) (`nano-banana-2-1`)

> Flagship 2K generative image engine with native prompt adherence and sharp typography

- **Modality**: `IMAGE`
- **Execution Provider**: `flow-agent`
- **Credit Cost**: 1 credits per run
- **Native Resolution**: 1024 × 1024
- **Max Resolution**: 2048 × 2048 (2K Master)
- **Supported Aspect Ratios**: 1:1, 16:9, 9:16
- **Average Latency**: ~0s

Our primary high-speed image synthesis engine powered by Google Flow. Supports native 1024x1024, 1376x768, and 768x1376 renders with real-time mathematical watermark inversion via gemini-omni.

#### Parameters:

- `prompt` (STRING): Describe what you want to synthesize in vivid detail
- `aspect_ratio` (ASPECT_RATIO): Aspect Ratio
- `resolution` (SELECT): Resolution

---

### Nano Banana Pro (`nano-banana-pro`)

> High-end editorial and studio advertising visual synthesis with sub-surface scattering

- **Modality**: `IMAGE`
- **Execution Provider**: `flow-agent`
- **Credit Cost**: 1.2 credits per run
- **Native Resolution**: 1024 × 1024
- **Max Resolution**: 2048 × 2048 (2K Master)
- **Supported Aspect Ratios**: 1:1, 16:9, 9:16
- **Average Latency**: ~0s

Tailored for editorial fashion, luxury product staging, and cinematic character closeups. Features advanced specular highlights and skin texture realism.

#### Parameters:

- `prompt` (STRING): Detailed aesthetic description
- `aspect_ratio` (ASPECT_RATIO): Aspect Ratio
- `resolution` (SELECT): Resolution

---

### Qwen-Image-2.1-Turbo (`qwen-image-2-1-turbo`)

> Ultra-fast multimodal foundation model with superior prompt obedience & multi-ref conditioning

- **Modality**: `IMAGE`
- **Execution Provider**: `huggingface`
- **Credit Cost**: 1 credits per run
- **Native Resolution**: 1024 × 1024
- **Max Resolution**: 2048 × 2048
- **Supported Aspect Ratios**: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3
- **Average Latency**: ~5s

Alibaba Qwen Image 2.1 Turbo foundation model. Delivers lightning-fast 4-step generation, photorealistic textures, and accurate spatial multi-concept synthesis from text and image references.

#### Parameters:

- `prompt` (STRING): Describe the scene, subject, styling, and lighting in detail
- `aspect_ratio` (ASPECT_RATIO): Aspect Ratio
- `steps` (NUMBER): Inference Steps
- `enhance` (SELECT): Prompt Enhancement

---

### FLUX.1 Schnell 12B (`flux-1-schnell`)

> 12-Billion parameter rectified flow transformer for instantaneous photorealistic synthesis

- **Modality**: `IMAGE`
- **Execution Provider**: `huggingface`
- **Credit Cost**: 1 credits per run
- **Native Resolution**: 1024 × 1024
- **Max Resolution**: 2048 × 2048
- **Supported Aspect Ratios**: 1:1, 16:9, 9:16, 4:3, 3:4
- **Average Latency**: ~4s

Black Forest Labs FLUX.1 Schnell model. Tailored for 1 to 4 step ultra-high-fidelity generation with unparalleled skin realism, typography, complex hand anatomy, and cinematic lighting.

#### Parameters:

- `prompt` (STRING): Describe subjects, environment, mood, and photographic lens specs
- `aspect_ratio` (ASPECT_RATIO): Aspect Ratio
- `num_inference_steps` (NUMBER): Inference Steps

---

### Google Veo 2 Cinematic (`google-veo-2`)

> High-definition temporal video synthesis with fluid camera dolly and coherent physics

- **Modality**: `VIDEO`
- **Execution Provider**: `flow-agent`
- **Credit Cost**: 10 credits per run
- **Native Resolution**: 1920 × 1080
- **Max Resolution**: 1920 × 1080 @ 24fps
- **Supported Aspect Ratios**: 16:9, 9:16
- **Average Latency**: ~22s

State-of-the-art cinematic text-to-video and image-to-video generator. Renders realistic fluid dynamics, atmospheric lighting, and accurate physical object interactions.

#### Parameters:

- `prompt` (STRING): Describe the scene, action, motion dynamics, and camera movement
- `aspect_ratio` (ASPECT_RATIO): Aspect Ratio

---

### Gemini-Omni (Lossless Watermark Inverter) (`gemini-omni`)

> Proprietary mathematical reverse-alpha watermark removal with zero blurring

- **Modality**: `ENHANCEMENT`
- **Execution Provider**: `flow-agent`
- **Credit Cost**: 0.2 credits per run
- **Native Resolution**: Lossless (Pixel-Exact)
- **Max Resolution**: 4096 × 4096
- **Supported Aspect Ratios**: ALL
- **Average Latency**: ~0s

Our proprietary Go core and SIMD AVX engine that mathematically inverts alpha-blended watermarks on images and videos. Unlike generic AI inpainting, it restores original pixel chrominance without blur or hallucination.

#### Parameters:

- `input_media` (STRING): Image or video with SynthID / Flow watermark to invert

---

### Flux 1.1 Pro (Black Forest Labs) (`flux-1-1-pro`)

> SOTA open-weights realism, anatomical precision, and complex scene composition

- **Modality**: `IMAGE`
- **Execution Provider**: `external-sota`
- **Credit Cost**: 1.5 credits per run
- **Native Resolution**: 1024 × 1024
- **Max Resolution**: 2048 × 2048
- **Supported Aspect Ratios**: 1:1, 16:9, 9:16, 4:3, 3:4
- **Average Latency**: ~5s

Industry standard for hyper-realistic human skin, natural hands, legible typography, and photorealistic lighting. Backed by 12B parameter flow matching transformer.

#### Parameters:

- `prompt` (STRING): Detailed prompt for Flux transformer
- `aspect_ratio` (ASPECT_RATIO): Aspect Ratio

---

### WAN 2.1 Video & Motion (`wan-2-1`)

> 14B parameter open-weights video foundation model for dynamic interaction & motion

- **Modality**: `VIDEO`
- **Execution Provider**: `self-hosted-gpu`
- **Credit Cost**: 8 credits per run
- **Native Resolution**: 1280 × 720
- **Max Resolution**: 1920 × 1080
- **Supported Aspect Ratios**: 16:9, 9:16, 1:1
- **Average Latency**: ~28s

Advanced open-weights video model capable of generating multi-subject dynamic physical interactions (hugging, kissing, dancing) from text or still reference photos.

#### Parameters:

- `prompt` (STRING): Motion Prompt
- `aspect_ratio` (ASPECT_RATIO): Aspect Ratio

---

### FaceFusion 3.0 Pro (`facefusion-3`)

> Ultra-precision 4K neural face swap with expression preservation and occluder masking

- **Modality**: `ENHANCEMENT`
- **Execution Provider**: `self-hosted-gpu`
- **Credit Cost**: 1.5 credits per run
- **Native Resolution**: Original Image Dimensions
- **Max Resolution**: 4096 × 4096
- **Supported Aspect Ratios**: ANY
- **Average Latency**: ~3s

SOTA face-swapping engine featuring InsightFace ArcFace embedding, facial landmark tracking, skin color tone matching, and automated CodeFormer post-enhancement.

#### Parameters:

- `source_face` (STRING): Image providing identity / face to insert
- `target_media` (STRING): Destination media where face will be swapped into

---

### BRIA RMBG 2.0 (Background Remover) (`bria-rmbg-2-0`)

> State-of-the-art background removal producing transparent PNGs with edge-level hair & sheer fidelity

- **Modality**: `ENHANCEMENT`
- **Execution Provider**: `huggingface`
- **Credit Cost**: 0.4 credits per run
- **Native Resolution**: Lossless Up to 4K
- **Max Resolution**: 4096 × 4096
- **Supported Aspect Ratios**: 1:1, 16:9, 9:16, 4:3, 3:4
- **Average Latency**: ~1s

BRIA RMBG 2.0 background matting model trained on commercial licensed datasets. Flawlessly segments fine hair strands, transparent glass, pets, e-commerce products, and portrait figures.

#### Parameters:

- `input_image` (STRING): Source image from which to extract subject and produce transparent alpha channel PNG

---

### BiRefNet Studio Matting (`birefnet-rmbg`)

> Sub-pixel background remover with hair strand and transparent object edge precision

- **Modality**: `ENHANCEMENT`
- **Execution Provider**: `self-hosted-gpu`
- **Credit Cost**: 0.3 credits per run
- **Native Resolution**: Lossless Resolution
- **Max Resolution**: 4096 × 4096
- **Supported Aspect Ratios**: ANY
- **Average Latency**: ~1s

Bilateral Reference Network (BiRefNet) for ultra high resolution image segmentation. Preserves individual strands of flying hair, sheer fabric translucency, and glass reflection.

#### Parameters:

- `input_image` (STRING): Image from which to strip the background

---

### CodeFormer 4K Face Restorer (`codeformer-restore`)

> Transformer-based facial restoration for repairing blurry faces, old portraits, and AI compression

- **Modality**: `ENHANCEMENT`
- **Execution Provider**: `huggingface`
- **Credit Cost**: 0.5 credits per run
- **Native Resolution**: Lossless Up to 4K
- **Max Resolution**: 4096 × 4096
- **Supported Aspect Ratios**: 1:1, 16:9, 9:16, 4:3, 3:4
- **Average Latency**: ~2s

SOTA face restoration algorithm leveraging codebook lookup priors to recover authentic facial landmarks, sharp irises, skin pores, and natural teeth without cartoonish smoothing.

#### Parameters:

- `input_image` (STRING): Image with faces to enhance and restore to high definition
- `codeformer_fidelity` (NUMBER): Fidelity Weight (0=Quality, 1=Identity)
- `upscale` (NUMBER): Upscaling Factor

---

### Depth Anything V2 (`depth-anything-v2`)

> Metric & relative monocular depth estimation for 3D stereoscopy, mesh reconstruction, and ControlNet

- **Modality**: `ENHANCEMENT`
- **Execution Provider**: `huggingface`
- **Credit Cost**: 0.3 credits per run
- **Native Resolution**: Lossless Up to 4K
- **Max Resolution**: 4096 × 4096
- **Supported Aspect Ratios**: 1:1, 16:9, 9:16, 4:3, 3:4
- **Average Latency**: ~1s

SOTA monocular depth estimation network trained on over 62 million images. Accurately segments foreground distance, subtle surface normals, and complex geometric foliage.

#### Parameters:

- `input_image` (STRING): Image from which to extract relative or metric depth map

---

### Idol 3D Gaussians (.PLY) (`idol-3d-gaussian`)

> One image in, a 3D object out. Transform any single photo into an interactive 3D Gaussian Splat (.PLY)

- **Modality**: `3D`
- **Execution Provider**: `huggingface`
- **Credit Cost**: 2 credits per run
- **Native Resolution**: Dense 3D Gaussian Splat
- **Max Resolution**: 1,048,576 Gaussians
- **Supported Aspect Ratios**: 1:1, 4:3, 3:4, 16:9
- **Average Latency**: ~9s

Idol by SPERIDLABS transforms a single photograph of an object into a rich, view-consistent 3D Gaussian Splat (.PLY). The resulting 3D point cloud can be rendered in real-time WebGL viewers or imported into Blender, Unreal Engine, and Unity.

#### Parameters:

- `input_image` (STRING): Single image of an isolated subject or object to reconstruct as a 3D Gaussian splat
- `mode` (SELECT): Prior optimization mode tailored for synthetic 3D assets or real-world photographs
- `steps` (NUMBER): Diffusion optimization iterations (higher values yield sharper geometry and colors)
- `cfg_scale` (NUMBER): Classifier-Free Guidance weight for geometry adherence
- `rescale_t` (NUMBER): Diffusion timestep trajectory rescaling
- `gaussians` (SELECT): Target count of 3D Gaussian primitives in the output .PLY point cloud

---