Skip to content
BranchesTags

Latest commit

 

History

26 Commits

Folders and files

NameName
Last commit message
Last commit date
configs
configs
 
 
scripts
scripts
 
 
utils
utils
 
 
.env.example
.env.example
 
 
.gitignore
.gitignore
 
 
.python-version
.python-version
 
 
AGENTS.md
AGENTS.md
 
 
CLAUDE.md
CLAUDE.md
 
 
LICENSE
LICENSE
 
 
README.md
README.md
 
 
requirements.txt
requirements.txt
 
 

Repository files navigation

Engines

This directory contains Python-based engines and batch processing scripts for the E-Dream GPU system.

Setup

  1. Install Dependencies: Ensure you have the required packages installed, including the edream_sdk.

    cd engines
pip install -r requirements.txt

  2. Environment Variables: Create a .env file in the engines directory (or use the one in the project root). BACKEND_URL and API_KEY are required. Deployment-specific UUIDs used as test inputs also live here; configs reference them with ${VAR} (e.g. "image_uuid": "${STILL_UUID}"). You can copy .env.example and replace the API_KEY.

    BACKEND_URL=https://api.infinidream.ai/api/v1
API_KEY=your_api_key_here
STILL_UUID=an-image-dream-uuid          # source for i2i / i2v scripts
DREAM_UUID=a-video-dream-uuid            # source for video scripts

Guidance / CFG

Each model family names this knob differently and accepts a different range:

Model Param Range Low / Mid / High
kling-i2v, kling-25-i2v cfg_scale 0–1 0 / 0.5 / 1
ltx-i2v guidance 1–5 1 / 3 / 5
wan-i2v guidance 0–10 0 / 5 / 10

The scripts reject an out-of-range value rather than letting the worker clamp it, and warn if the config uses the other model's param name. The configured value is stamped into each dream's name (e.g. [cfg_scale=0.5]) so results from several runs into one playlist stay identifiable, and it is part of the resume key so re-running the same prompt at a different value is not skipped as already-done.

Batch Processing Scripts

These scripts are located in engines/scripts/ and use the edream_sdk to interact with the API directly.

1. Wan Image-to-Video Batch (run_wan_i2v_batch.py)

Generates videos from image dreams using the Wan I2V algorithm with various prompt combinations. Source images come from an image_playlist_uuid (typically the output playlist of run_qwen_image_batch.py or run_z_image_turbo_batch.py), or from a single image_uuid.

Configuration (engines/configs/job.json):

{
    "image_playlist_uuid": "${STILL_PLAYLIST_UUID}",
    "prompt": "A cinematic shot of...",
    "combos": ["in a cyberpunk city", "underwater"],
    "playlist_uuid": "optional-existing-playlist-uuid",
    "playlist": {
        "name": "My Batch Videos",
        "description": "Generated from batch script",
        "nsfw": false
    },
    "size": "1280x720",
    "duration": 5,
    "num_inference_steps": 30,
    "guidance": 5
}

Guidance is guidance, range 0–10 (default 5).

Usage:

python3 scripts/run_wan_i2v_batch.py

2. Uprez Batch — removed

run_uprez_batch.py has been removed. It predated the backend's native uprez_playlist algorithm and hand-built the derived dreams, which produced a playlist that could not be re-run, could not requeue failed dreams, could not be cancelled as a unit, and had no keyframes. It also appended a tracking marker to each source dream's description, corrupting descriptions that carry real metadata.

Use python-api/scripts/uprez_playlist.py instead. It creates a derived playlist whose own prompt names the source, and lets the backend materialize and link the dreams:

python scripts/uprez_playlist.py <source_playlist_uuid> \
    --upscale-factor 2 --interpolation-factor 2

3. Qwen Image Batch (run_qwen_image_batch.py)

Generates multiple images from a prompt and downloads them locally.

Configuration (engines/configs/qwen-image-config.json):

{
    "prompt": "A futuristic cityscape...",
    "num_generations": 5,
    "output_folder": "generated_images",
    "size": "1024x1024",
    "seed": -1
}

Example (use an existing playlist):

{
    "prompt": "A futuristic cityscape...",
    "num_generations": 5,
    "output_folder": "generated_images",
    "size": "1024x1024",
    "seed": -1,
    "playlist_uuid": "existing-playlist-uuid"
}

Example (create a new playlist):

{
    "prompt": "A futuristic cityscape...",
    "num_generations": 5,
    "output_folder": "generated_images",
    "size": "1024x1024",
    "seed": -1,
    "playlist": {
        "name": "Qwen Image Batch",
        "description": "Generated from qwen image batch script",
        "nsfw": false
    }
}

Usage:

python3 scripts/run_qwen_image_batch.py

4. LTX Image-to-Video Batch (run_ltx_i2v_batch.py)

Generates videos from a playlist of image dreams using LTX 2.3.

Configuration (engines/configs/ltx-i2v-config.json):

Single image:

{
    "image_uuid": "${STILL_UUID}",
    "prompt": "A cinematic shot of...",
    "duration": 5,
    "guidance": 1,
    "seed": -1,
    "playlist": { "name": "LTX I2V Output", "nsfw": false }
}

Batch from playlist:

{
    "image_playlist_uuid": "source-image-playlist-uuid",
    "prompt": "A cinematic shot of...",
    "combos": ["in a cyberpunk city", "underwater"],
    "playlist_uuid": "optional-existing-playlist-uuid",
    "playlist": { "name": "LTX I2V Batch Output", "nsfw": false },
    "duration": 5,
    "seed": -1,
    "lora": "ltx-2-19b-lora-camera-control-static.safetensors",
    "lora_strength": 0.4
}

Guidance is guidance, range 1–5 (default 1). LTX uses a distilled few-step sampler tuned near 1.0, so higher values change the image far less than on Wan.

Usage:

python3 scripts/run_ltx_i2v_batch.py

5. Disco Diffusion Batch (run_disco_batch.py)

Generates images or animations using Disco Diffusion v5.2 (CLIP-guided diffusion with optional optical-flow warp).

Configuration (engines/configs/disco-config.json):

Single image:

{
    "batch_name": "DiscoTest",
    "animation_mode": "None",
    "width": 512,
    "height": 512,
    "steps": 50,
    "skip_steps": 10,
    "clip_guidance_scale": 5000,
    "cutn_batches": 1,
    "clip_vit_b32": true,
    "seed": 1337,
    "text_prompts": {
        "0": ["A beautiful painting of a cosmic nebula, trending on artstation"]
    },
    "playlist": { "name": "Disco Output", "nsfw": false }
}

2D animation (dream zoom):

{
    "batch_name": "DreamZoom",
    "animation_mode": "2D",
    "max_frames": 24,
    "fps": 6,
    "zoom": "0:(1.02)",
    "angle": "0:(0)",
    "frames_skip_steps": "70%",
    "steps": 50,
    "clip_guidance_scale": 3000,
    "cutn_batches": 1,
    "clip_vit_b32": true,
    "seed": 42,
    "text_prompts": {
        "0": ["A cosmic nebula, trending on artstation, by Greg Rutkowski"]
    },
    "playlist": { "name": "Disco Output", "nsfw": false }
}

Video Input (optical-flow stylization of a source video):

{
    "batch_name": "VideoWarp",
    "animation_mode": "Video Input",
    "flow_warp": true,
    "flow_blend": 0.5,
    "check_consistency": true,
    "source_dream_uuid": "your-source-video-dream-uuid",
    "steps": 50,
    "clip_guidance_scale": 3000,
    "cutn_batches": 1,
    "clip_vit_b32": true,
    "text_prompts": {
        "0": ["A painting in the style of Van Gogh, trending on artstation"]
    },
    "playlist": { "name": "Disco Output", "nsfw": false }
}

Animation modes: None (single image), 2D (zoom/rotate/translate), 3D (depth-warped), Video Input (optical-flow warp)

Usage:

python3 scripts/run_disco_batch.py

6. Nvidia VSR Batch — removed

run_nvidia_vsr_batch.py has been removed for the same reason as run_uprez_batch.py above: it hand-built the derived dreams, so the playlists it produced had no source_dream_uuid linking them to their sources and could not be re-run, requeued, cancelled as a unit, or played continuously.

Use python-api/scripts/uprez_playlist.py with the VSR algorithm. RTX VSR is resolution-only, so it takes no interpolation factor and its quality is the NVIDIA enum (LOW/MEDIUM/HIGH/ULTRA), not the x264 preset:

python scripts/uprez_playlist.py <source_playlist_uuid> \
    --dream-algorithm nvidia-uprez --upscale-factor 2 --quality ULTRA

6. Z-Image Turbo Batch (run_z_image_turbo_batch.py)

Generates images using the Z-Image Turbo model. Supports text-to-image and image-to-image generation.

Configuration (engines/configs/z-image-turbo-config.json):

{
    "prompt": "a beautiful landscape with mountains and a lake",
    "num_generations": 2,
    "output_folder": "generated-images",
    "output_filename": "z-image-turbo",
    "size": "1024*1024",
    "seed": -1,
    "output_format": "png",
    "enable_safety_checker": true,
    "playlist": {
        "name": "Z-Image Turbo Batch",
        "description": "Generated from z-image-turbo batch script",
        "nsfw": false
    }
}

For image-to-image, add image (URL) and optionally strength (0.0–1.0):

{
    "prompt": "a futuristic version of this scene",
    "image": "https://example.com/input.jpg",
    "strength": 0.8,
    "output_format": "jpeg"
}

Valid size options: 512*512, 768*768, 1024*1024, 1280*1280, 1024*768, 768*1024, 1280*720, 720*1280

Valid output_format options: png, jpeg, webp

Usage:

python3 scripts/run_z_image_turbo_batch.py

7. FLUX Schnell Batch (run_flux_schnell_batch.py)

Text-to-image generation with FLUX.1 [schnell] (fal). Downloads results locally.

Configuration (engines/configs/flux-schnell-config.json):

{
    "prompt": "A vibrant sunset over ocean waves, photorealistic.",
    "num_generations": 2,
    "output_folder": "generated-images",
    "output_filename": "flux-schnell",
    "size": "1280*720",
    "num_inference_steps": 4,
    "seed": -1,
    "playlist": { "name": "FLUX Schnell Batch", "nsfw": false }
}

Valid size options: 1024*768, 1024*1024, 768*1024, 1280*720, 720*1280

Usage:

python3 scripts/run_flux_schnell_batch.py

8. FLUX Kontext Image-to-Image Batch (run_flux_kontext_i2i_batch.py)

Re-imagines source images with an edit prompt using FLUX.1 Kontext (fal). Output follows the source size, so there is no size parameter. Sources come from a single image_uuid or an image_playlist_uuid.

Configuration (engines/configs/flux-kontext-i2i-config.json):

{
    "image_uuid": "${STILL_UUID}",
    "prompt": "Turn it into a watercolor painting with soft pastel colors.",
    "seed": -1,
    "output_folder": "generated-images",
    "output_filename": "flux-kontext",
    "playlist": { "name": "FLUX Kontext I2I Batch", "nsfw": false }
}

Usage:

python3 scripts/run_flux_kontext_i2i_batch.py

9. Kling Image-to-Video Batch (run_kling_i2v_batch.py)

Animates source images with a prompt using Kling (fal). Set model to kling-i2v (Kling 3.0 Pro) or kling-25-i2v (Kling 2.5 Turbo Pro). Video results are auto-uploaded to their dream, so nothing is downloaded locally.

Configuration (engines/configs/kling-i2v-config.json):

{
    "model": "kling-i2v",
    "image_uuid": "${STILL_UUID}",
    "prompt": "The scene comes alive with gentle, cinematic motion.",
    "duration": 5,
    "negative_prompt": "",
    "cfg_scale": 0.5,
    "playlist": { "name": "Kling I2V Batch", "nsfw": false }
}

Durations: kling-i2v allows 3–15s; kling-25-i2v allows 5s or 10s. Optionally set end_source_uuid for an end frame.

Guidance is cfg_scale, range 0–1 (default 0.5). Kling does not accept a guidance key — setting one is ignored, and the scripts warn when they see it.

Usage:

python3 scripts/run_kling_i2v_batch.py

About

use the python api to create playlists from prompts

Resources

Readme
AGPL-3.0 license
Activity
Custom properties

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages