AI RenderScope – User Manual

A desktop viewer for inspecting image metadata, prompts and generation parameters.

Figure 1 – Main window. Clicking the preview opens an external, resizable full‑resolution window.

1. Description

The AI RenderScope is a viewer-only application. It lets you browse folders of images, read their positive and negative prompts, and inspect metadata such as generator, model, sampler, scheduler, LoRA modules, seed, steps, CFG, and noise values. The app does not generate new images or re-run filters. Its goal is to make audit, curation and QC of datasets fast and reliable.

Clicking the in-window preview opens a separate external window that shows the selected image at its full resolution. You can move that window to another monitor and resize it for side-by-side review.

2. Interface Overview

2.1 Path to the Images (left panel)

Shows drives and folders on your computer.
Navigate to the dataset directory. Selecting a folder lists its images.

2.2 Image Preview (center)

Displays the currently selected image.
Use the navigation arrows above the preview to move through images in the folder.
Click the preview to open the external full-resolution window.

2.3 Prompt Section (bottom-center)

Positive Prompt – Tags and descriptors intended to appear in the image.
Negative Prompt – Elements to avoid or suppress in generation.

Contains dropdowns for Generator, Model, Sampler, Scheduler and LoRA. These controls are for viewing and organizing metadata only in this viewer.

2.5 Info Section (right)

Generator: Backend used to create the image (e.g., InvokeAI, ComfyUI).
Model: The model or checkpoint that produced the image.
Sampler: The sampling algorithm. Different samplers yield different textures and stability.
Scheduler: Step scheduling strategy used by some pipelines.
Seed: The random seed. Same seed + same parameters usually reproduce the same composition.
Steps: Number of denoising steps. Higher values can add detail, but with diminishing returns.
CFG: Classifier-Free Guidance scale. Higher means stronger adherence to the prompt, but too high may reduce naturalness.
Noise: Initial noise strength or offset if present.
LoRA: List of LoRA adapters and their weights, e.g., girl_gg1:1.0.

3. How to Use

Load images – In the left panel, browse to a folder and click an image filename.
Inspect the preview – Use the arrows to navigate. Check the on-screen prompts.
Open full resolution – Click the preview to open the external window and zoom visually by resizing that window.
Review metadata – The right panel shows generator, model, sampler, scheduler, seed, steps, CFG, noise, and LoRA data.

4. Parameter Reference

Generator

The tool or UI that orchestrated image creation. Examples: InvokeAI, ComfyUI, Automatic1111. Knowing the generator helps you replicate pipelines and samplers correctly.

Model

The base or fine-tuned model used to synthesize the image, e.g., blobForTraining_v10. Versioning matters for visual consistency and reproducibility.

Sampler

The diffusion sampler controls how noise is iteratively removed. Common samplers include Euler, DPM++ 2M SDE K, etc. Different samplers influence sharpness, coherence and speed.

Scheduler

Defines the per-step noise schedule. Not all pipelines expose it. Pairing sampler and scheduler consistently helps match a specific look.

Seed

A numeric value that initializes the random process. Reusing the same seed with identical parameters typically reproduces the same layout and pose.

Steps

The number of denoising iterations. Typical ranges are 20–40. Too low can cause artifacts, too high increases render time with minimal gains.

CFG (Guidance)

Strength of guidance toward the positive prompt versus the unconditional prior. Moderate values like 6–12 usually balance prompt fidelity and realism.

Noise

Initial noise strength or a specific noise offset value, when recorded. Useful to understand domain differences across pipelines.

LoRA

Lightweight adapters that steer the base model toward new concepts. The notation name:weight indicates which LoRAs were active and how strongly.

5. Tips and Best Practices

Use the external window to examine fine details at native resolution.
Record prompt conventions and keep them consistent across datasets.
When comparing images, note sampler and CFG together. Changes there often explain stylistic shifts.
Keep an audit trail of model and LoRA versions to reproduce results later.

6. Notes and Limitations

This application is a viewer. It does not generate or modify images. Dropdowns in the filter panel are informational and may be used for organization or search in future versions.

If some metadata fields are missing, they were not embedded in the image or could not be parsed from sidecar data.