Sampler Settings
The Sampler Settings panel in AIFX allows for detailed control over the technical aspects of your image generation. Adjusting these lets you fine-tune image quality, style, and performance.
IMAGE SOURCE
Definition:
The optional starting image for img2img (image-to-image) generation. Instead of generating from just a text prompt, img2img allows you to provide an initial source image that the AI will transform based on your prompt and settings. The denoising process begins with your chosen source, enabling creative variations or style transfers from an existing picture.
TIP
- Your source image must have width and height divisible by 8 for best results and compatibility.
INFO
- AIFX will automatically resize your image in the latent space when sampling if needed, or you can manually resize it within the software.
RES
Controls the overall resolution of your generated image – both width and height. Choosing the right resolution balances image detail and memory usage.
Both width and height must be divisible by 8 (e.g., 512, 768, 1024).
TIP
For even cleaner, artifact-free images, try making your dimensions divisible by 64—for example, 768 × 1152.
- Clicking on the The Res button activates a dropdown menu that provides easy access to common and recommended resolutions.
PIPELINE
Definition:
The pipeline determines the processing method used to generate your image. Each option is optimized for different models or workflows. Some pipelines use quantized models (indicated by a "Q" in their name), which require less GPU memory and are ideal for use on systems with less powerful graphics cards.
Common Pipeline Options:
- FLUX_DEV: Standard pipeline for Flux models.
- FLUX_SCHNELL: Optimized for faster generation with Flux models.
- SDXL: Designed for Stable Diffusion XL and related high-resolution models.
- SD3.5: For the Stable Diffusion 3.5 series of models.
TIP
Choose the pipeline that matches your selected model. If working with limited hardware resources, look for quantized variants (ending in "Q") for improved performance and memory efficiency.
MODEL
Definition:
The AI model you use to generate art. Each model can have a distinct "look" and capabilities.
Example:Model: TheLounger/Jib_Mix_Flux_v7_Beta-GGUF
GUIDANCE
Definition:
This is the "CFG" (Classifier-Free Guidance) value, controlling how strongly the generation follows your prompt (higher means more literal, lower means more creative).
- Flux Models: Guidance acts as real CFG—affecting prompt adherence and creativity balance.
- 1.0 is no CFG
- 2.0 usually brings a strong result
- SD 3.5: Typical CFG setting is
5.0(balanced adherence and creativity).
Try starting with the model’s recommended range (e.g., 5–7 for Flux, 5 for SD 3.5), and experiment up or down for your ideal blend of control versus creative freedom.
STEPS
Definition:
Controls how many processing steps are used to generate your image. More steps typically mean better quality, but longer render times.
Recommended Values:
- Turbo pipelines (like FLUX_SCHNELL, SD3.5 Turbo): 4–12 steps
- Standard pipelines (SD3.5, Flux): 30–60 steps
- SDXL: Up to 60–100 steps
TIP
Some models and pipelines are designed for turbo-speed rendering (like FLUX_SCHNELL or SD3.5 Turbo) and work well with very few steps. If you’re using standard pipelines, start at 30–60 steps for quality results, and increase as needed.
Start lower for quick tests, and increase for final renders.
NOISE
Definition:
Controls how much random noise is added to the image each step as it is denoised into a coherent image.
- Lower noise: More simple plainer images with familiar shapes.
- Higher noise: More unexpected, chaotic, or experimental results.
Example:Noise: 0.5
TIP
If you set noise to 1.0, your image may become "crazy" or chaotic—try starting at 0.5, and if it's too wild, cut it in half, rerun, and inch upward for the sweet spot.
TIP
- Use dimensions divisible by 8 (or 64 for extra).
- For creativity: Try lower guidance (CFG), higher noise.
- For coherent, detailed art: Increase steps, keep noise moderate, and use propper guidance.
Experiment with combinations to find your unique style!