v0.7.0

🚀 Deferred Diffusion v0.7.0

Docker Images

docker pull joegaffney/deferred-diffusion:api-0.7.0
docker pull joegaffney/deferred-diffusion:worker-0.7.0
docker pull joegaffney/deferred-diffusion:agentic-0.7.0
docker pull joegaffney/deferred-diffusion:comfy-0.7.0

Release Package

Download the release archive below for clients and docker compose files.

To run

1. Download and extract the archive:

tar -xzf deferred-diffusion-0.7.0.tar.gz

2. Change to the extracted directory:

cd deferred-diffusion

3. Start the services using Docker Compose:

docker-compose down
docker-compose pull
docker-compose up -d --no-build

• See README.md and DEPLOYMENT.md in the archive for more details.
• For troubleshooting or advanced configuration, refer to the documentation in the repository.

What's Changed

• feature: refactors model naming conventions {remove vendor prefix) and introduces ComfyUI workflow support to nuke client. by @JoeGaffney in #59
• feature: improved API key management and identity tracking on tasks by @JoeGaffney in #60
• feature: improved API key handling and identify based logging. by @JoeGaffney in #61
• feature: changes the output delivery mechanism from base64-encoded data to file storage with signed URLs by @JoeGaffney in #63
• feature: task metadata exposure via Flower API integration, consolidates helper functions, and reorganizes output file storage paths by @JoeGaffney in #64
• feature: simplifying the release process, sronger admin key - updated docs. by @JoeGaffney in #65

Full Changelog: v0.6.0...v0.7.0

v0.6.0

🚀 Deferred Diffusion v0.6.0

Docker Images

docker pull joegaffney/deferred-diffusion:api-0.6.0
docker pull joegaffney/deferred-diffusion:worker-0.6.0
docker pull joegaffney/deferred-diffusion:comfy-0.6.0

Release Package

Download the release archive below for clients and docker compose file.

Note: The included docker-compose.yml contains the core API and worker services. ComfyUI workflow integration is available but requires manual configuration - see the main repository's docker-compose.yml for reference.

To run

1. Download and extract the archive:

tar -xzf deferred-diffusion-0.6.0.tar.gz

2. Change to the extracted directory:

cd deferred-diffusion

3. Start the services using Docker Compose:

docker-compose down
docker-compose pull
docker-compose up -d --no-build

• See README.md in the archive for more details.
• For troubleshooting or advanced configuration, refer to the documentation in the repository.

What's Changed

• wip: sam-3 video by @JoeGaffney in #51
• feature: logging that enables progress tracking of task execution across image and video generation by @JoeGaffney in #52
• feature: support for the Hunyuan Video 1.5 model and refactors resolution handling across the codebase by @JoeGaffney in #53
• feature: workflows end point for calling sidecar comfy (experimental feature) by @JoeGaffney in #54
• refactor: removed adapters and control nets from sd-xl and flux-1 by @JoeGaffney in #55
• feature: improved comfy sidecar integration with more robust client by @JoeGaffney in #56
• feature: refactors the ComfyUI sidecar integration to be optional and externally configurable, adds support for video workflows by @JoeGaffney in #57

Full Changelog: v0.5.0...v0.6.0

v0.5.0

🚀 Deferred Diffusion v0.5.0

Versioned Docker Images

docker pull joegaffney/deferred-diffusion:api-0.5.0
docker pull joegaffney/deferred-diffusion:worker-0.5.0

Latest Docker Images

docker pull joegaffney/deferred-diffusion:api-latest
docker pull joegaffney/deferred-diffusion:worker-latest

Release Package

Download the release archive below for clients and docker compose file.

To run

1. Extract the archive:

2. Change to the extracted directory:

cd deferred-diffusion

3. Start the services using Docker Compose:

docker-compose down
docker-compose pull
docker-compose up -d --no-build

• See README.md in the archive for more details.
• For troubleshooting or advanced configuration, refer to the documentation in the repository.

What's Changed

• chore: Tidied up README and Deployment info. Removed unused configuration vars by @JoeGaffney in #49
• feature: z-image, flux-2-pro. Updated houdini and nuke clients. Some schema simplification by @JoeGaffney in #50

Full Changelog: v0.4.0...v0.5.0

v0.4.0

🚀 Deferred Diffusion v0.4.0

Versioned Docker Images

docker pull joegaffney/deferred-diffusion:api-0.4.0
docker pull joegaffney/deferred-diffusion:worker-0.4.0

Latest Docker Images

docker pull joegaffney/deferred-diffusion:api-latest
docker pull joegaffney/deferred-diffusion:worker-latest

Release Package

Download the release archive below for clients and docker compose file.

To run

1. Extract the archive:

2. Change to the extracted directory:

cd deferred-diffusion

3. Start the services using Docker Compose:

docker-compose down
docker-compose pull
docker-compose up -d --no-build

• See README.md in the archive for more details.
• For troubleshooting or advanced configuration, refer to the documentation in the repository.

What's Changed

• feature: domain specific task names. added sam-3, flux-2. Moved some runway models to replicate provider by @JoeGaffney in #46
• feature: text_encoding in the main pipeline doing on CPU was using too much RAM and was too slow by @JoeGaffney in #47
• feature: consolidated memory size and handling offloading. Removed direct precision control by @JoeGaffney in #48

Full Changelog: v0.3.0...v0.4.0

v0.3.0

🚀 Deferred Diffusion v0.3.0

Versioned Docker Images

docker pull joegaffney/deferred-diffusion:api-0.3.0
docker pull joegaffney/deferred-diffusion:worker-0.3.0

Latest Docker Images

docker pull joegaffney/deferred-diffusion:api-latest
docker pull joegaffney/deferred-diffusion:worker-latest

Release Package

Download the release archive below for clients and docker compose file.

To run

1. Extract the archive:

2. Change to the extracted directory:

cd deferred-diffusion

3. Start the services using Docker Compose:

docker-compose down
docker-compose pull
docker-compose up -d --no-build

• See README.md in the archive for more details.
• For troubleshooting or advanced configuration, refer to the documentation in the repository.

What's Changed

• feature: improved agentic layers state - added task cancelation by @JoeGaffney in #42
• wip: task status enum - aligning client implementations by @JoeGaffney in #43
• refactor: made more parity in task names to module names for improved clarity. Consolidated some runway tasks by @JoeGaffney in #44
• feature: wide ranging update with system prompt, models, and clients by @JoeGaffney in #45

Full Changelog: v0.2.0...v0.3.0

v0.2.0

🚀 Deferred Diffusion v0.2.0

Docker Image

docker pull joegaffney/deferred-diffusion:api-0.2.0
docker pull joegaffney/deferred-diffusion:worker-0.2.0

Release Package

Download the release archive below for clients.

What's Changed

• feature: WAN vace first pass. Update pytorch image. Moved dynamic model router to tests by @JoeGaffney in #34
• feature: Split API endpoints into /local/{model} and /external/{model} variants by @JoeGaffney in #35
• refactor: video pipeline rename image_last_frame to last_image and switch WanVACE to video-to-video workflow by @JoeGaffney in #36
• feature: Added new Minimax Hailuo-2, splitting monolithic test files into separate files per model by @JoeGaffney in #38
• feature: it tests support both local and external API testing with improved CI/CD integration by @JoeGaffney in #39
• revert: going back to unified endpoints as simplier for MPC and clients by @JoeGaffney in #40
• feature: unify model metadata management - MCP server by @JoeGaffney in #41

Full Changelog: v0.1.0...v0.2.0

v0.1.0

🚀 Deferred Diffusion v0.1.0

Docker Image

docker pull joegaffney/deferred-diffusion:api-0.1.0
docker pull joegaffney/deferred-diffusion:worker-0.1.0

Release Package

Download the release archive below for clients.

What's Changed

• feature: fast api and pytest by @JoeGaffney in #1
• feature: generated client by @JoeGaffney in #2
• fix: spec operation ids and added test_data by @JoeGaffney in #3
• Update README.md by @JoeGaffney in #4
• feature: ip adapter support by @JoeGaffney in #5
• feature: modular control nets and cache / memory optimisations by @JoeGaffney in #6
• feature: added scale layers choice to ip_adapaters by @JoeGaffney in #7
• feature: sequence agent by @JoeGaffney in #8
• feature: controlnet and adapter updates for shot workflows by @JoeGaffney in #9
• feature: pluralized structure, IPAdapter and ControlNet exception handling and new models by @JoeGaffney in #10
• feature: stateless file data using base64, update video models, docker build by @JoeGaffney in #11
• feature: nuke integration by @JoeGaffney in #12
• feature: event driven architecture with workers by @JoeGaffney in #13
• feature: fixed up the HDA's to use the new event workflow by @JoeGaffney in #14
• feature: openai vison text calls - updated to use Makefile by @JoeGaffney in #15
• feature: infra AWS cloud dev kit by @JoeGaffney in #16
• feature: Auth, auto quant, hi dream and runway image models by @JoeGaffney in #17
• feature: nuke toolkit and video models tweaks by @JoeGaffney in #18
• feature: initial comfy workflow integration by @JoeGaffney in #19
• feature: improved schemas and split workers tasks to process, workflow and external by @JoeGaffney in #20
• feature: comfy video workflow and various improvements by @JoeGaffney in #21
• feature: this PR removes ComfyUI workflow support, adds Flux Kontext as a new external model option by @JoeGaffney in #22
• feature: simplified image models to best in class local and external - to avoid model zoo by @JoeGaffney in #23
• feature: cpu and gpu workers - basic release mechanism by @JoeGaffney in #24
• feature: added video nodes to nuke added LOW_VRAM to support videos on sub 16gb workers by @JoeGaffney in #25
• feature: nuke text nodes - improved UX - WAN fixes by @JoeGaffney in #26
• feature: updated to latest runway models - added image upscaler's by @JoeGaffney in #27
• This PR represents a major update to the system, upgrading model versions and significantly refactoring the codebase by @JoeGaffney in #28
• feature: modernizing the codebase from family-based task routing to direct model-specific task handling. by @JoeGaffney in #29
• feature: PR replaces older AI video generation models with newer ones and standardizes video/image resolution handling by @JoeGaffney in #30
• feature: Nunchaku-optimized transformers for Flux and Qwen image pipelines, align model naming/mappings by @JoeGaffney in #31
• feature: mypy ran on workers, redundant code cleaned up, high_quality toggle by @JoeGaffney in #32
• feature: CI/CD automation, better client organization, and documentation updates by @JoeGaffney in #33

New Contributors

• @JoeGaffney made their first contribution in #1

Full Changelog: https://github.com/JoeGaffney/deferred-diffusion/commits/v0.1.0