Flatten repo structure for ComfyUI Manager
This commit is contained in:
Fred LeFevre
@@ -1,131 +0,0 @@
|
||||
# FSL Gemini & Veo Workflow for ComfyUI
|
||||
|
||||
**Version:** 6.5 (The "Production" Edition)
|
||||
**Powered By:** Google DeepMind (Gemini 1.5/2.0/3.0, Veo 3.1, Imagen 3)
|
||||
|
||||
This node suite creates a full-stack, cloud-powered AI media studio inside ComfyUI. It offloads heavy processing to Google Cloud, allowing for high-fidelity **Text-to-Video**, **Image-to-Video**, and **Agent-Assisted Creation** without requiring massive local VRAM.
|
||||
|
||||
---
|
||||
|
||||
## ✨ Core Features
|
||||
|
||||
1. **The "Creative Director" Agent:** A memory-enabled Chat Node that acts as a middleman. It intelligently detects if you want to chat or create. If you want to create, it writes professional "Hidden Hooks" to control the generators.
|
||||
2. **True Image-to-Video (Identity Preservation):** Direct integration with **Veo 3.1** allows you to send local pixels to the cloud. Veo animates *your* specific image rather than hallucinating a new one from text.
|
||||
3. **Smart Gatekeeping ($0 Cost Mode):** The Image and Video nodes feature "Gatekeeper Logic." If you are just chatting with the Agent (e.g., "How are you?"), the generators receive empty prompts and "sleep" instantly, incurring **zero cost**.
|
||||
4. **Robust Cloud Polling:** Uses a custom HTTP polling engine to bypass Python SDK timeouts, ensuring large video files download successfully even if generation takes minutes.
|
||||
5. **Hybrid Logic:** Automatically switches Veo configuration based on input type (Text vs. Image) to prevent API validation errors.
|
||||
|
||||
---
|
||||
|
||||
## 🛠️ Installation
|
||||
|
||||
### 1. Requirements
|
||||
These nodes rely on Google's modern Unified SDK (`google-genai`).
|
||||
|
||||
**For ComfyUI Portable (Windows):**
|
||||
Open a terminal in your `ComfyUI_windows_portable` folder and run:
|
||||
```bash
|
||||
.\python_embeded\python.exe -m pip install --upgrade google-genai requests
|
||||
```
|
||||
|
||||
**For Standard Python:**
|
||||
```bash
|
||||
pip install --upgrade google-genai requests
|
||||
```
|
||||
|
||||
### 2. API Key & Billing
|
||||
* **Gemini/Imagen:** Works with standard Google AI Studio keys.
|
||||
* **Veo (Video):** Requires a Google Cloud Project with **Billing Enabled**. Free-tier keys usually return 403/404 errors for video.
|
||||
|
||||
**Setup:**
|
||||
* **Option A (Best):** Set an Environment Variable `GEMINI_API_KEY` or `GOOGLE_API_KEY` in your startup batch file.
|
||||
* **Option B:** Paste the key directly into the `api_key` widget on every node.
|
||||
|
||||
---
|
||||
|
||||
## 📦 The Node Suite
|
||||
|
||||
### 1. FSL Gemini Chat (Unified SDK)
|
||||
* **Role:** The Director / Logic Engine.
|
||||
* **Inputs:** Text Prompt + Optional Image (Vision).
|
||||
* **Agent Mode:**
|
||||
* **Text-to-Media:** If you ask for a video, it writes a detailed prompt (Lighting, Camera Move, Physics) into the `MEDIA_PROMPT_HOOK`.
|
||||
* **Image-to-Media:** If an image is connected, it writes a "Fidelity Prompt" instructing the generator to *preserve the identity* of the input image.
|
||||
* **Outputs:** `LATEST_REPLY` (Chat), `MEDIA_PROMPT_HOOK` (Technical Prompt), `CONVERSATION_HISTORY` (Log).
|
||||
|
||||
### 2. FSL Google Veo Generator
|
||||
* **Role:** The Animator.
|
||||
* **Inputs:** Text Prompt (from Chat Hook) + Optional Image.
|
||||
* **Feature:** **Save Location**.
|
||||
* *Note:* Paths are relative to the **ComfyUI Root Folder**, not the Output folder.
|
||||
* Example: `my_videos` saves to `ComfyUI/my_videos/`.
|
||||
* Example: `output/my_videos` saves to `ComfyUI/output/my_videos/`.
|
||||
* Example: `C:/Videos/` saves to an absolute path.
|
||||
|
||||
### 3. FSL Imagen Generator (Pro)
|
||||
* **Role:** The Artist (Fast).
|
||||
* **Engine:** Google Imagen 3.
|
||||
* **Gatekeeper:** Returns a tiny black square instantly if prompt is empty.
|
||||
|
||||
### 4. FSL Gemini Generate Image (Legacy)
|
||||
* **Role:** The Editor (Advanced).
|
||||
* **Usage:** Keep this node if you need **Inpainting**, **Masking**, or **Image-to-Image** editing features not present in the lightweight Pro node.
|
||||
* *Tip:* If you encounter timeouts with this node, edit the `.py` file and increase `timeout=60` to `timeout=180`.
|
||||
|
||||
---
|
||||
|
||||
## 🚦 Veo Logic & Modes (Critical)
|
||||
|
||||
The **Veo Generator** node has two distinct modes of operation depending on what you connect to it.
|
||||
|
||||
### Mode A: Image-to-Video (Identity Preservation)
|
||||
* **Trigger:** A wire is connected to `image_input`.
|
||||
* **Behavior:** Veo animates the pixels provided.
|
||||
* **Constraint:** **Duration and Aspect Ratio sliders are IGNORED.**
|
||||
* *Why?* The Veo 3.1 API currently crashes if you try to force a custom duration or aspect ratio onto an existing image. The node automatically strips these settings to prevent errors.
|
||||
* *Result:* You will get the default Veo output (usually ~5 seconds, Native Aspect Ratio).
|
||||
|
||||
### Mode B: Text-to-Video (Generation)
|
||||
* **Trigger:** Nothing is connected to `image_input`.
|
||||
* **Behavior:** Veo generates a video from scratch based on words.
|
||||
* **Capabilities:** Full control over `duration_seconds` (5s - 8s) and `aspect_ratio`.
|
||||
|
||||
### 🔄 How to Switch Modes
|
||||
If you want to create a video from scratch (Text-to-Video), you **must disconnect** the image wire.
|
||||
* **Method:** Drag the wire off the `image_input` pin.
|
||||
* If you leave the image connected but type "Create a video of a spaceship", Veo will get confused and try to turn your input image *into* a spaceship (or crash).
|
||||
|
||||
---
|
||||
|
||||
## 🔌 Wiring Strategies
|
||||
|
||||
### Strategy A: The "Creative Director" (Text-to-Video)
|
||||
1. Connect `FSL Gemini Chat` (Hook) -> `FSL Veo Generator` (Prompt).
|
||||
2. Ensure `image_input` on Veo is **Disconnected**.
|
||||
3. **Prompt:** "Generate a cinematic video of a cyberpunk city."
|
||||
|
||||
### Strategy B: "Identity Preservation" (Image-to-Video)
|
||||
1. **The Y-Split:** Connect your **Load Image** node to **TWO** inputs:
|
||||
* `FSL Gemini Chat` -> `image_input` (So the Agent sees context).
|
||||
* `FSL Veo Generator` -> `image_input` (So Veo gets the raw pixels).
|
||||
2. **Prompt:** "Animate this person turning their head."
|
||||
3. **Result:** The Agent tells Veo to "preserve fidelity," and Veo uses the direct pixel connection to animate your exact image.
|
||||
|
||||
---
|
||||
|
||||
## ❓ Troubleshooting
|
||||
|
||||
**Q: Veo Error: "The number value for durationSeconds is out of bound"**
|
||||
* **A:** This happens if you are in **Image-to-Video** mode but older code was trying to force a duration. Ensure you are using **v6.5+** of the nodes, which automatically disables config parameters when an image is detected.
|
||||
|
||||
**Q: Load Video Path Error: "video is not a valid path..."**
|
||||
* **A:** This means the Veo node failed to generate the MP4, so it passed an Error Message string instead of a file path. Check the ComfyUI console/terminal to see the real error (usually API Key or Safety Filter related).
|
||||
|
||||
**Q: "RAI Media Filtered" / Blocked by Safety**
|
||||
* **A:** Google blocked the generation. Veo is strict about photorealistic people. Try describing the *action* rather than the *person*, or try the "Minimize Safety" toggle (though Veo 3.1 is very strict regardless).
|
||||
|
||||
---
|
||||
|
||||
## 📝 Credits
|
||||
* **Developer:** FSL (Fred's Super Logic)
|
||||
* **Engine:** Google DeepMind (Gemini / Veo / Imagen)
|
||||
Reference in New Issue
Block a user