€10

Character Prompt Catalogue | + Products | Comfyui Workflow | 1.2

1 rating
I want this!

Character Prompt Catalogue | + Products | Comfyui Workflow | 1.2

€10
1 rating

Advanced ComfyUI Workflow with Gemini 2.0 Beta: Character & Product Consistency

Introduction

This guide presents a professional-grade workflow for ComfyUI, based on Gemini 2.0 Flash Experimental (Beta), designed to generate high-quality AI images with consistent characters and reliable product integration. This system is ideal for creative, advertising, and editorial contexts where you need to maintain the same talent across dozens of variations. The documentation outlines two primary workflows: one focused on character consistency, and the other on character + product integration in realistic scenarios. Both flows can be activated independently and also support manual configuration, offering maximum flexibility for intermediate or advanced users.

Workflow Overview

This workflow includes two core functions, each of which can be executed individually:

1. Character Consistency

Starting from a reference photo of the talent (e.g., a front portrait), the system generates 50–100 consistent images of the same subject, varying in pose, expression, camera angle, and scenery. However, the core facial and identity features remain identical across all images. This makes the workflow suitable for applications like fashion campaigns, storyboarding, character design, and more.

In addition to producing visually coherent variations, this process also enables the easy creation of LoRA training sets, as the resulting images maintain such a high level of consistency and reliability. Each image is generated through either batch-injected prompts or manually configured prompts via .txt files. The core engine for generation is the Gemini 2.0 Flash Exp Image Generation node.

2. Character + Product

This flow accepts a reference image of the talent plus one or more images of the product and/or environment, and outputs 50–100 images that show the person using, interacting with, or posing alongside the product in a variety of realistic scenes. Outputs may include commercial photoshoots, advertising visuals, still life shots with models, and more.

Like the character-only flow, this scenario also uses batch or manual prompt injection and is built on top of the Gemini 2.0 model. However, it differs primarily by the number and type of image inputs: where the character-only flow uses one input image, this flow uses multiple.

Prompt Generation GPTs:

Prompt Structure and Categories

Each GPT is designed to output a structured set of prompts grouped into creative categories. This ensures that the generated images span multiple dimensions of variation—angle, emotion, composition—while preserving the subject’s identity.

Character Prompt Catalogue

This GPT focuses exclusively on the talent and produces about 60–70 prompts organized in five key categories:

  • ROTATION (20 prompts)
    Explore the subject from multiple angles (front, profile, 3/4 view, back, etc.) while maintaining consistent visual features (face, outfit, lighting, expression).
  • EXPRESSION (20 prompts)
    Show different emotional states—smiles, intense gazes, anger, sadness, surprise—without changing hairstyle, clothing, lighting, or core pose.
  • FRAMING / CAMERA ANGLES (10 prompts)
    Change the camera perspective: close-up, wide-angle, bird’s-eye view, low angle, etc. The subject remains constant, only the perspective changes.
  • AMBIENTATION & ATMOSPHERE (10 prompts)
    Place the character in various environments. Atmosphere may vary (light, fog, color, temperature), but the character remains visually consistent.
  • OTHER / NOT CLASSIFIED (10 prompts)
    Creative prompts outside standard categories: subtle costume variations, symbolic poses, micro-narratives, or object interactions.

Character Prompt Catalogue + Products

This GPT combines the character with the product and returns prompts structured into seven categories:

  • ROTATION (20 prompts)
    Views rotating around the character and product to showcase silhouettes, textures, volumes, and poses.
  • EXPRESSION (20 prompts)
    Facial and body language variations that reflect emotion, mood, or storytelling—while integrating the product into the scene.
  • FRAMING / CAMERA ANGLES (10 prompts)
    Specific camera setups (close-up, worm’s eye, over-the-shoulder) to highlight the product design and the presence of the character.
  • AMBIENTATION & ATMOSPHERE (10 prompts)
    Environments with matching emotional tones (nighttime, urban, ethereal, domestic, futuristic), ensuring harmony between person and product.
  • OTHER / NOT CLASSIFIED (10 prompts)
    Visually and narratively impactful prompts that don’t fall into standard categories.
  • ADS (10 prompts)
    Advertising-style prompts with strong visual storytelling designed for conversion. Perfect for mockups, portfolios, or branded content.
  • STILL LIFE PRODUCT ONLY (10 prompts)
    Product-only scenarios focusing on materials, textures, composition, and usage for campaigns, displays, or e-commerce.

⚠️ GPT output is progressive: it generates blocks of prompts first, and then compiles a full table ready for .txt file use at the end.

es.
Your job is to render the character facing directly forward with a neutral, composed posture. Maintain all other elements — masculine facial structure, wavy black hair, light olive skin with freckles, trimmed beard, intense green eyes, soft studio lighting, and a white shirt under a dark blazer.

Your job is to show the character in a quarter-left rotation, looking slightly down with a thoughtful expression. Maintain all other elements — masculine facial structure, wavy black hair, light olive skin with freckles, trimmed beard, intense green eyes, soft studio lighting, and a white shirt under a dark blazer.

Your job is to render the character in a quarter-right turn, showing a soft smile. Maintain all other elements — masculine facial structure, wavy black hair, light olive skin with freckles, trimmed beard, green eyes, soft studio lighting, and a white shirt under a dark blazer.

Your job is to depict the character in full profile facing left, with a reflective look. Maintain all other elements — masculine facial structure, wavy black hair styled with volume, olive-freckled skin, well-groomed beard, green eyes, soft studio lighting, and a white shirt under a dark blazer.

Your job is to depict the character in profile facing right with a stern expression. Maintain all other elements — masculine facial structure, wavy black hair styled with natural volume, olive-freckled skin, trimmed beard, green eyes, soft studio lighting, and a white shirt under a dark blazer.

Your job is to show the character from behind, turning his head over his shoulder toward the camera. Maintain all other elements — wavy black hair with tousled volume, light olive skin with freckles, beard, soft studio lighting, and the white shirt and dark blazer combination.

Your job is to render the character with his head tilted upward, chin raised confidently. Maintain all other elements — masculine facial structure, voluminous black hair, olive skin with freckles, beard, green eyes, soft studio lighting, and white shirt under a dark blazer.

Your job is to depict the character tilting his head downward, casting subtle shadows over his eyes. Maintain all other elements — wavy black hair, light olive freckled skin, beard, green eyes, soft studio lighting, and tailored clothing.

Preparing the Prompt Files (.txt)

Once the prompt lists are generated by the GPTs (or written manually), they must be loaded into the workflow using plain text files:

  1. Prompt Creation/Editing: When using the GPTs, request output in simple plain text format (.txt)—avoiding any table formatting—so that the prompts can be copied easily. Alternatively, you may manually write your own list, with each line containing one full prompt (covering all desired elements like pose, style, scene, etc.).
  2. Creating .txt Files: Create two separate text files:
    • Character.txt — for prompts related to character-only generation.
    • Character+Product.txt — for prompts combining character and product.
    Copy the relevant prompt lines into the respective files and save them.
  3. Placing Files in the Input Directory: Move these .txt files into the appropriate input folder within ComfyUI, typically ComfyUI/input/:
    ComfyUI/
 └── input/
      ├── Character.txt
      └── Character+Product.txt
    Ensure the filenames match what the ComfyUI graph expects. If different names are used, update the file loader node paths accordingly.
  4. Format Check: Each line in the file will be read as a distinct prompt. Avoid blank lines between prompts. Commenting (using # or //) may be supported depending on the file loader node used, but it's safer to keep only usable lines.

Once saved and placed correctly, the workflow can batch-generate images using these prompts. Alternatively, individual prompts can still be manually inserted into the workflow interface for more controlled testing.

Executing the Workflow in Batch Mode

After models are loaded and prompt files are prepared, you’re ready to run batch image generation:

  1. Load the Workflow: Open ComfyUI and load the workflow JSON file. You can do this by dragging the JSON file into the interface or using the Load button. You will see a node graph that includes sections for image input, prompt input, model loading (Flux, Gemini, etc.), and output.
  2. Provide Reference Images: Locate the Gemini 2.0 Flash Experimental node in the graph. It accepts two image inputs:
    • Image 1 — Reference photo of the talent (typically a well-lit frontal portrait).
    • Image 2 — Reference image of the product and/or environment, required for the "Character + Product" flow. Leave blank or use a placeholder image if running only the character flow.
    Tip: Use high-quality, square (1024×1024) images for optimal results with Gemini. The product image should clearly showcase the item’s main features.
  3. Select the Desired Flow: Ensure that the appropriate text file node is linked in the graph:
    • For Character Only, use Character.txt.
    • For Character + Product, use Character+Product.txt.
    In the workflow file, both file reader nodes are included and can be switched by toggling their connections. You may also manually update the file path inside the node if needed.
  4. Start Batch Generation: Press Execute (▶️) in ComfyUI to start generating. The workflow is set up to iterate through each line of the selected text file:
    • A counter node increases the index for each execution.
    • A text file loader fetches the corresponding line.
    • That prompt is sent to the Gemini node with the image references.
    • The output image is saved in the ComfyUI output/ folder.
    Repeat the generation until all lines from the .txt file are used. You can queue multiple runs manually or use ComfyUI’s built-in Queue if available.
  5. Monitor and Stop: Check images as they are generated to verify facial consistency and prompt interpretation. Once the final prompt is reached, stop the execution. Running past the end of the file may result in blank or unexpected outputs.
  6. Manual Mode (Optional): If you prefer manual entry, disconnect the prompt loader and write prompts directly into the Gemini node. This mode is ideal for testing or minor edits.

✅ Tip: For repeated generation, you can reset the counter node using its “reset” input, allowing you to restart from the top of the prompt list without restarting ComfyUI.

Model and Module Requirements

To operate properly, the workflow depends on a set of specific models and custom nodes within ComfyUI. Below is the full list of required components along with installation instructions.

Gemini 2.0 Flash Experimental (API-based)

  • Function: Google’s experimental multimodal model for high-consistency image generation based on reference images. It excels at maintaining identity and style coherence across multiple outputs.
  • Access: This is not a downloadable model. You must obtain an API Key through Google AI (https://aistudio.google.com/) or the Gemini developer preview.
  • Integration:
    • Install the ComfyUI Gemini Plugin: https://github.com/CY-CHENYUE/ComfyUI-Gemini-API
    • Place it in your ComfyUI/custom_nodes/ directory.
    • After restarting ComfyUI, insert the Gemini node into your graph, input your API key, and select gemini-2.0-flash-exp as the model.
    • Choose aspect ratio, temperature, and other generation parameters as needed.

Flux 1.0 [dev] (UNet Model)

Flux 1.0 Redux [dev] (Style Model)

Text Encoders

  • T5-XXL (for rich prompt interpretation)
  • OpenCLIP-Large (CLIP-L)
    • File: clip_l.safetensors
    • Same HuggingFace repo as above
    • Place in ComfyUI/models/text_encoders/

VAE (Variational Autoencoder)

LoRA Models

These lightweight add-ons fine-tune output quality and detail:

All LoRAs should be placed in ComfyUI/models/loras/. The workflow uses Power LoRA Loader by rgthree for toggling multiple LoRA files.

Upscaler (Face Focused)

Additional ComfyUI Extensions Required

Ensure all custom nodes are placed in ComfyUI/custom_nodes/ and restart ComfyUI after installation. Once all components are in place, your system is ready for advanced generation with consistency across characters and product imagery.

Final Recommendations & Optimization Tips

Once all models and extensions are correctly installed, your system is ready for advanced, consistent, and high-volume image generation. Here are a few final tips to help optimize your workflow:

  • Output Quantity Control: The number of generated images corresponds to the number of prompt lines in the .txt file. You can adjust this by editing the file content or switching between different prompt files. Reset the index counter node to restart batch generation if needed.
  • Execution Time: Gemini operates via cloud API, so generation time may vary (typically a few seconds per image). Producing 100 images might take several minutes depending on network latency and image complexity.
  • Prompt Editing During Execution: You can edit the .txt file mid-process. The loader node will fetch the updated prompt as it reaches that line. This is useful for quickly correcting or refining prompt behavior.
  • Product Fidelity: If a product's appearance changes across images, enhance prompt descriptions by explicitly stating key traits (e.g., “a blue cylindrical bottle with a silver cap”) to maintain consistency.
  • API Safety Filters: Gemini applies safety filters. Avoid inappropriate content (nudity, violence, etc.) to prevent errors or blurred outputs.
  • Upscaling Strategy: Use the included upscaler (4xFaceUpDAT) for final render passes. You may disable it for faster previews and re-enable it once the images are finalized.

This workflow provides a robust system for automated character and product image generation, with creative control, high consistency, and flexibility. Perfect for agencies, content studios, e-commerce brands, and AI creators working with identity-driven visuals

Guided Exercise Example

Inside the ZIP file you downloaded, you'll find 5 key files to get you started:

  • Anna.jpg – Your AI talent reference image
  • COCO MADEMOISELLE.jpg – The product to be used
  • character.txt – Pre-written prompts for character consistency
  • character+product.txt – Pre-written prompts for character + product scenes
  • Character Prompt Catalogue _ + Products _ ComfyUI Workflow.json – The full ComfyUI workflow file

Use the Fast Groups Bypasser node in ComfyUI to activate only the modules you intend to work with.

🔧 IMPORTANT:

  • 01 Flux Model → This module must always be active.

If you're using automated prompt loading:

  • 02 Load Prompt from File → enable this node
  • Then choose one of the following (never both):
    • 03 Gemini single image
    • 05 Gemini multi image

If you're using manual prompt input instead:

  • Use one of these, based on the image input type:
    • 04 Gemini single image manual prompt
    • 06 Gemini multi image manual prompt

📌 REMEMBER:

  • Configure the Temperature setting in the Gemini module for more or less creative results.
  • For upscaling, feel free to tweak the parameters until you reach the best visual output for your needs.
  • Ensure stable internet connection for Gemini API.

This hands-on setup will help you get started with generating highly consistent character and product visuals using the provided assets.

Important Update

If you’ve correctly configured the Gemini API and the Google service at
https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash-exp-image-generation:generateContent
returns an HTTP/1.1 400 Bad Request, the issue is not related to the workflow, but rather to the API itself.

As of 2:00 PM on April 7, 2025, the service is experiencing problems.
We are currently investigating whether Google is transitioning all calls to paid-only access.

Stay tuned for further updates.


🔄 Update 1.2 – Multi-Image Prompt Control: Pose & Outfit Matching

This update introduces advanced multi-image prompting capabilities within the Gemini 2.0 + Flux1 ComfyUI workflow. You can now generate high-consistency character images that closely replicate poses and/or outfits from additional reference images.

What’s New

Using the multi-image input module in the workflow, you can now generate:

  • Pose-matched variations, based on the second, third, or fourth reference image you upload
  • Pose + outfit matched compositions, by referencing a full-body photo of your model with the desired styling

Available Prompt Tools

  • Pose-Based Prompt Generator
    Generates prompts that replicate the subject’s body orientation, hand placement, leg posture, and spatial flow based on a reference pose image. The generator keeps the facial structure and styling consistent with the primary character while adapting body language and gesture.


  • Pose + Outfit Prompt Generator
    Generates prompts that replicate both the physical stance and clothing style from a full-body reference photo. This allows for fashion-consistent outputs across multiple scenarios, ideal for catalog shoots, editorial simulations, and try-on experiences.

How to Use

  • Prepare your prompt .txt file as usual (e.g., character+product.txt)
  • Upload your primary face reference as Image 1
  • Upload additional pose and/or outfit images as Image 2 (multi-reference input)
  • Select the Gemini multi-image node path in ComfyUI
  • Run your batch as usual to generate 50–100 pose/outfit-matched results

I want this!

A professional ComfyUI workflow for generating highly consistent character and product images using Gemini 2.0. Includes automated prompt batching, multi-image input support, and ready-to-use assets for immediate deployment.

Size
1.28 MB
Copy product URL
30-day money back guarantee

Ratings

5
(1 rating)
5 stars
100%
4 stars
0%
3 stars
0%
2 stars
0%
1 star
0%