PRINTFILM: Self-Hosted AI Video and Manga Creation Platform

What PRINTFILM Does and Who It Targets

PRINTFILM is an open-source platform for AI-generated content production. Its two product lines cover AI manga (animated comic episodes) and AI short videos. The pipeline runs from a user-supplied theme or script through storyboard editing, AI image generation, AI video generation, and FFmpeg-based final composition to a playable file. The README describes the result as "能播的片子" (a ready-to-broadcast episode).

The platform targets short-video producers, marketing teams who want to turn product pitches into playable clips, and manga creators who need consistent character and scene assets across multiple episodes. Voiceover is generated by Seedance during video production, with no separate dubbing step. The use cases listed in the README include short videos for customer acquisition, episodic manga from outlines, standalone image or video generation, and self-hosted deployment for teams who want to keep media on their own infrastructure.

Version 0.2.0 was released on 2026-09-17. The repository is MIT licensed.

Two Product Lines: AI Manga and AI Short Video

The AI manga line handles the full production of episodic content. A user starts with an outline or plot summary, which the system parses into a script, then into individual shots. The README references docs/EPISODE_RULES.md for the specific rules governing episode parsing. Characters, scenes, and props go into a reusable asset library, which maintains visual consistency across episodes.

The AI short video line uses pre-built style templates and a linear pipeline. The README states there are 20 or more built-in style templates. Two output modes are available: `full`, which generates images, video, and the composite final file; and `image_text`, which produces static images only, described as faster and cheaper. Individual shots can be redrawn or regenerated without restarting the entire pipeline. Tasks continue running if a user leaves the page.

A Tool Center provides standalone generation capabilities outside the full pipeline: text-to-image, image-to-image, image-to-product, text-to-video, video-to-video, and e-commerce collage.

Deploying PRINTFILM with Docker

The repository ships a public Docker image hosted on Alibaba Cloud Container Registry under the `gcc` namespace. No login is required to pull the image. The README states that only Docker Desktop (or Docker Engine with Compose) is needed on the host machine.

To start, clone the repository and copy the example environment file:

git clone https://github.com/yi1108/printfilm.git
cd printfilm
cp deploy/.env.docker.example deploy/.env.docker

Edit deploy/.env.docker to set at minimum the database password, a session secret key, and an API key. Then start the stack:

docker compose --env-file deploy/.env.docker up -d

After roughly 30 seconds for the PostgreSQL health check to pass, four services are available: the user frontend at http://localhost:8080, the admin backend at http://localhost:8081, the API and its Swagger documentation at http://localhost:8000 and /docs, and a health check endpoint at http://localhost:8000/api/health.

To check container status, follow logs, or stop the stack:

docker compose --env-file deploy/.env.docker ps
docker compose --env-file deploy/.env.docker logs -f api
docker compose --env-file deploy/.env.docker down

A separate docker-compose.full.yml file is available for building the image from source when the public registry image is inaccessible or when local code changes need to be included.

Configuring the TokenFree API for Real Generation

The open-source release uses TokenFree New API for all AI operations: text (LLM), image generation, and video generation. Without a valid API key, the platform can only run in mock mode (ARK_MOCK=true), which shows the interface but does not produce real content.

To enable real generation, add the key to deploy/.env.docker:

OPENAI_API_KEY=sk-your-key
OPENAI_BASE_URL=https://www.tokenfree.com/v1
ARK_API_KEY=sk-your-key
ARK_MOCK=false
MODEL_LLM=kimi-k2.6
MODEL_IMAGE=seedream-5-0-pro
MODEL_VIDEO=seedance-2-5

Both OPENAI_API_KEY and ARK_API_KEY take the same key. After editing the file, recreate the API container to apply the environment:

docker compose --env-file deploy/.env.docker up -d --force-recreate api

The health endpoint at http://localhost:8000/api/health shows an `ark_mock` field that should read `false` once configured. The admin backend at http://localhost:8081 provides an alternative path: open System Settings, navigate to Models, and enter the key in the TokenFree channel. Keys stored through the admin UI are encrypted in the database.

In-Process Task Platform and Technical Architecture

The backend runs on Python 3.12 with FastAPI and SQLAlchemy. The database is PostgreSQL 16; the cache and task queue layer uses Redis 7. The frontend is React 19 with TypeScript and Vite 8, and the admin interface uses Tailwind CSS with shadcn components.

Rather than requiring a separate Celery worker process, PRINTFILM uses an in-process task platform with a scheduler, executor, and poller. The README describes this as a deliberate design choice that reduces deployment complexity. The docker-compose.yml confirms the architecture: PostgreSQL and Redis are separate containers, but the API container handles task execution internally.

The model used for each role (LLM, image, video) is configurable through environment variables or the admin backend, without code changes. The admin backend also handles user management, order management, template management, and model routing. An open API at /api/v1 accepts Bearer token or X-Api-Key authentication for image and video generation.

Limitations: TokenFree Dependency and Geographic Scope

All AI generation in the default configuration flows through TokenFree New API. Teams outside China should verify that the service is accessible from their network and that the required models (kimi-k2.6, seedream-5-0-pro, seedance-2-5) are available on their account. The README does not document how to substitute a different API provider; the models listed are specific to the TokenFree endpoint.

The Docker image is hosted on Alibaba Cloud Container Registry's Hangzhou region (gcc-registry.cn-hangzhou.cr.aliyuncs.com). Pull speeds will vary for users outside East Asia, and while the image is listed as public, its availability for teams in specific regions should be tested before committing to this deployment path.

Billing is disabled by default. When enabled, the README directs users to docs/BILLING.md for the billing and Yipay (易支付) configuration. Teams operating outside of China should evaluate whether the billing integration is compatible with their payment infrastructure before enabling it.

The video output from Seedance includes voiceover generation, but the README does not specify which languages or voices are supported for that narration.

Comparison with ComfyUI: Vertical Studio versus Node-Based Workflow

ComfyUI is a widely used open-source, self-hosted tool for building AI image and video generation workflows using a visual node graph. It supports many of the same underlying model types for image and video generation, and it can be extended to cover complex multi-step pipelines.

The difference is in the level of the abstraction. ComfyUI requires building and connecting nodes to define a pipeline; a script-to-storyboard-to-video workflow in ComfyUI would need to be constructed manually. PRINTFILM provides that pipeline as a ready-made web application with a dedicated interface for script editing, shot management, asset libraries, and episode continuity. For a creative team that wants to produce content without configuring node graphs, PRINTFILM offers a more direct path. For a technical team that wants to customize every generation step or experiment with different models and pipelines, ComfyUI is more flexible.

ComfyUI does not include built-in project management for multi-episode anime productions, voiceover generation from a model, or a billing system.

Editorial conclusion

Teams building short-form video or manga production pipelines who want full control over media and credentials will find PRINTFILM a complete self-hosted foundation. The dependency on TokenFree New API for all AI generation means teams outside China should verify service availability before deploying. Billing is disabled by default but can be enabled through docs/BILLING.md. The last push was on 2026-09-24.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. Project website
4. README
5. yi1108/printfilm on GitHub