Sign In
homemodelsimagesvideospostsarticlesbountieschallengeseventsupdatesshop

How to Generate a Photoshoot Scene from Multiple Angles with a Stable Subject in Automatic1111

11

SA

sarahpeterson

Mar 12, 2025

workflows
stable charactermultiple scenesfixed seedmultiple viewsworkflow
How to Generate a Photoshoot Scene from Multiple Angles with a Stable Subject in Automatic1111

How to Generate a Single Photoshoot Scene from Multiple Angles with a Stable Subject in Automatic1111

Creating a consistent scene across multiple angles in AI-generated imagery can be challenging, especially when aiming for a stable subject and environment. This guide walks through a streamlined method using Automatic1111’s WebUI, leveraging combinatorial generation , fixed seeds, dynamic prompts, and LoRAs to achieve professional photoshoot-style results.

Prerequisites

  1. Automatic1111 WebUI installed (with the Dynamic Prompts extension enabled).

  2. A LoRA (Low-Rank Adaptation) model trained for subject/scene consistency (e.g., a character-specific or style-specific LoRA).

  3. Basic familiarity with prompt engineering and batch processing.

Step-by-Step Workflow

1. Set a Fixed Seed & Scene Anchor

  • Why a Fixed Seed? A fixed seed ensures the base noise pattern remains consistent across all generated images, stabilizing the subject and environment.

    • In the Dynamic Prompts tab:

      • Enable Fixed Seed (under "Advanced" settings).

  • Anchor the Scene in the Prompt:
    Start your prompt with unchanging details to lock the subject, location, and style. For example:

    Copy

    A young woman with curly red hair, standing in a futuristic cityscape, wearing a neon-lit jacket, <lora:affelejFT15c:0.9> affelejftc, 1girl, realistic, hetero, 1boy, intricate details, 8k

    • Replace <lora:affelejFT15c:0.9> with your LoRA trigger.

2. Use a LoRA for Subject Consistency

  • LoRAs trained on specific subjects or styles help maintain coherence across angles.

    • Add the LoRA trigger to your prompt (e.g., <lora:affelejFT15c:0.9>).

    • Adjust the LoRA weight (e.g., :0.8) if needed to balance influence.

3. Add Dynamic Shot Types via Wildcards

  • Insert a wildcard or piped syntax at the end of your prompt to vary camera angles:

    Copy

    {wide angle|POV|close-up|overhead shot|low angle}, cinematic lighting

    • This generates one unique shot type per batch image.

4. Configure Combinatorial generation

  • In the Batch Count/Batch Size settings:

    • Set Batch Size to the number of shot types (e.g., 5 for the example above).

    • If using multiple LoRAs, use Batch count instead as we cant have multiple lora in one batch.

  • Enable Combinatorial generation and set max generations (Dynamic Prompts extension) to iterate through all shot types.

5. Generate and Refine

  • Click Generate and review the output.

    • Ensure the subject, clothing, and background remain stable.

    • Troubleshoot inconsistencies by:

      • Increasing the LoRA weight.

      • Simplifying the scene description.

      • Adjusting the CFG scale (7–12 recommended).

Example Workflow

Prompt:

Copy

__spps_dcl__,
22 year old, {supermodel, very skinny, petite,|||} {__breastsize__, __jumbo/appearance/haircolour__ __hair-female__, {__jumbo/people/nationalities/europe__ real world location | caucasian}, {__skin-color__ skin|||}, __eyecolor__ eyes, fair skinned, | hair, breasts, eyes,}  best quality , RAW photo, subject, 8k uhd,  high quality, photorealistic, { {__luxury__   |},   __luxury__  , {__sarahpeterson/scene_fashion__,||||}  skinned male, 
{__detail__  ,  |}       <lora:add_detail:1>, |||}   __luxury__,  {__sarahpeterson/*__,||||} __sarahpeterson/scene_mansion__, __jumbo/medium/photography/*__, __sarahpeterson/shot/*__,

Settings:

  • Seed: Fixed in dynamic prompts.

  • Batch Count: 4 (matching the four shot types).

  • Combinatorial generation

  • Combinatorial batches: 1

  • Max generation : 4 (set to batch count).

  • Sampler: DPM++ 2M Karras.

  • Steps: 30.

Tips for Success

  1. Use High-Quality LoRAs: Train or download LoRAs specifically designed for consistency.

  2. Balance Specificity: Avoid overloading the prompt; keep scene details concise but vivid.

  3. Test Seeds: Experiment with different seeds to find one that aligns with your vision.

Common Pitfalls

  • Inconsistent Subjects: Fix the seed and increase LoRA weight if features drift.

  • Overcomplicated Prompts: Too many variables can destabilize the scene.

By combining fixed seeds, LoRAs, and dynamic shot types, you can generate cohesive multi-angle scenes perfect for storytelling, concept art, or virtual photoshoots. Experiment with different prompts and LoRAs to refine your results!

11

Comments