logo
Docs

Workflows & Models

OpenFork workflows are Docker-backed GPU services. The website, desktop app, Python client, and cloud images all read from the same service registry so model availability, VRAM requirements, and routing stay aligned.

62

Enabled services

100

Workflow entries

2GB

Lowest VRAM

80GB

Highest VRAM

Source of truth: service metadata lives in services.json. Workflow processors and workflow JSON live in the Python client.

Video

Text-to-video, image-to-video, video-to-video, and native audio-video generation.

daVinci MagiHuman

daVinci-MagiHuman through Wan2GP with Magi Human Distill SR1080 quanto int8 weights. Use for realistic portrait talking-head/lip-sync shots only. Requires a realistic portrait start image; generates synchronized speech/audio from the text prompt.

16-32GB180-220GB disk

DomainShuttle

DomainShuttle subject-driven image-to-video using Wan2.2-DomainShuttle-A14B; registered as an 80GB service because the closest published upstream requirement is Wan2.2 A14B inference

80GB420GB disk

DreamID-Omni

DreamID-Omni FP8 talking-head video with identity and voice reference inputs

24GB140GB disk

LTX-2.3 Official

Official LTX-2 ltx-pipelines two-stage text-to-video generation plus image-to-video generation with the LTX-2.3 22B dev checkpoint and a trained LoRA package. This is separate from the LTX-2.3 Wan2GP GGUF services.

16-32GB220-240GB disk

SCAIL-2

Experimental SCAIL-2 14B WanGP candidate using the DeepBeepMeep int8 SCAIL-2 weights, SAM3 Magic Mask assets, and the native scail2_14B WanGP model type. Use for OpenFork 16GB smoke testing before production runs.

16-24GB220GB disk

Vista4D

Vista4D 384p49 through Wan2GP for source-video novel-view reshooting with predefined camera trajectories.

24GB200GB disk

WAN 2.2

High-quality video (Text/Image-to-Video)

8-24GB160-220GB disk

Image

Text-to-image, instruction editing, inpainting, and illustration models.

Anima

Anima text-to-image illustration model

8-16GB140GB disk

ERNIE-Image

Baidu ERNIE-Image Turbo text-to-image with CPU offload for 8GB GPUs

8-24GB90-120GB disk

FLUX Kontext

FLUX.1 Kontext [dev] GGUF Q4_K_M for 8GB VRAM; supports optional two-image composition by precomposing the source and identity reference

8-24GB120-140GB disk

Ideogram 4

Ideogram 4 text-to-image using NF4 weights with CPU offload and structured JSON layer prompts; true 16GB Vast smoke passed on 2026-06-09

16-32GB110-130GB disk

Krea 2

Experimental Krea 2 Turbo GGUF Q3 text-to-image via ComfyUI-GGUF with Qwen3-VL text encoder and Qwen Image VAE

8-24GB120-140GB disk

PiD

NVIDIA PiD 4x image super-resolution using the Flux/Z-Image-compatible VAE decoder path.

16GB80GB disk

Qwen Edit

Qwen-Image-Edit-2511 instruction editing plus Qwen-Image-2512 character LoRA text-to-image inference

8-12GB120-150GB disk

Qwen LoRA

Qwen-Image-2512 character LoRA text-to-image inference using the Unsloth Q4_K_M GGUF diffusion model for 24GB GPUs

24GB120GB disk

Qwen Turbo

Ultra-fast instruction-based editing with an optional second reference image (2 steps)

8GB120GB disk

TeleStyleV2

TeleStyleV2 content/style reference transfer on Qwen-Image-Edit-2509 with the TeleStyle and DMD LoRAs; upstream target is H100 80GB

80GB320GB disk

Z-Image

High-quality Z-Image (Q4_K_M GGUF)

8-24GB120-160GB disk

Z-Image Turbo

ControlNet & Flux Image Gen

8GB120GB disk

Audio

TTS, voice clone/design, music, sound effects, foley, and speech restoration.

ACE-Step 1.5

ACE-Step 1.5 XL music generation for 16GB VRAM

8-16GB100GB disk

AudioX

AudioX text-to-audio and video-conditioned sound generation

16-24GB120GB disk

Chatterbox

High-quality voice cloning

8GB80GB disk

DiffRhythm

AI-powered music composition

8GB100GB disk

dots.tts MF

Rednote HiLab dots.tts MeanFlow-distilled continuous TTS and zero-shot voice cloning, smoke tested on the 6GB image

6GB70GB disk

HeartMuLa

4-bit quantized music generation (~8GB VRAM)

16-24GB100GB disk

MMAudio V2A

Video-to-audio synthesis (MMAudio small_44k)

8-16GB80-100GB disk

PRiSM Audio

High-fidelity video-to-audio generation (PRiSM)

8-16GB70GB disk

Qwen3-TTS

Alibaba's multilingual TTS with 9 premium voices

8GB100GB disk

Qwen3-TTS 1.7B

Alibaba's multilingual TTS with voice design (1.7B)

16GB120GB disk

Scenema Audio

XML-driven expressive speech generation with zero-shot voice cloning

16GB160GB disk

Speech Enhancement

High-efficiency speech restoration and enhancement

2GB60GB disk

Stable Audio 3

Stable Audio 3 Small-SFX sound-effects-only text-to-audio

2GB40GB disk

VibeVoice

Clone-ready text-to-speech

8GB80GB disk

Utilities

Upscaling, prompt helpers, and local utility services used by production workflows.

LLM

Local Qwen3 4B LLM for workflow planning on low-VRAM providers

4GB20GB disk

LTX-2.3 Official

Official Lightricks LTX-2 Trainer package with the LTX-2.3 22B dev checkpoint. Standard upstream training targets 80GB+ VRAM; this lane targets the upstream 32GB low-VRAM INT8 config. 24GB is not an official supported trainer target.

24-32GB220GB disk

SparkVSR

State-of-the-art video super-resolution (24GB VRAM required)

24GB70GB disk

How Jobs Run

Submit

The website creates a DGN job with workflow type, inputs, storage target, policy, and billing plan.

Route

The orchestrator chooses a capable provider, preferring cached images and policy-compatible providers.

Process

The Python client pulls the Docker image if needed, runs the processor, uploads outputs, and settles credits or paid earnings.

Job Policies

Choose private, trusted, public, or paid routing