Choosing a sampler and a scheduler

<!-- lesson: samplers -->

Lesson "samplers", in the group Pictures: models of the lesson list. Steps: 10.

#What you will be able to do

You put two samplers side by side that differ in one setting only, and learn to choose by comparing.

#What you need

• It goes on from A first picture from text. The card says so; it is a hint, not a lock.
• Nodes: Load Checkpoint, CLIP Text Encode (Prompt), Empty Latent Image, KSampler, VAE Decode, Save Image / Preview Image, Preview Image / Save Image.
• The app downloads and installs nothing. The step texts below say which files and folders, and which ComfyUI template names them.

#The steps

The words of each step, as the lesson shows them. Node names are in bold.

1. Two settings, one way to choose (reading)

   The sampler (sampler_name) is the way each step removes noise. The scheduler (scheduler) decides how much noise is left before each step. Together they change the look of the picture, its detail and how many steps it needs.

   There is no best pair for every model. The way to choose is to compare: the same seed, the same steps and cfg, one setting changed. This lesson builds such a comparison on one canvas.

2. A picture graph is needed

   This lesson goes on from the graph of the lesson "A first picture from text". The canvas must have Load Checkpoint, two CLIP Text Encode (Prompt) nodes, Empty Latent Image, KSampler, VAE Decode and Save Image with their links.

   If they are not there, do that lesson first.

3. Add a second sampler

   Add a second KSampler node and connect to it the same as to the first:

   • to model, the output of Load Checkpoint;
   • to positive and negative, the same two text nodes;
   • to latent_image, the same Empty Latent Image.

   Both samplers start from one empty latent and one prompt.

4. Make everything the same

   In KSampler set seed, steps, cfg and scheduler the same as in the first sampler.

   The seed is the number the starting noise is made from. With the same seed both start from the same noise, so any difference in the result comes from the one setting you change.

5. Change only the sampler

   In KSampler choose another sampler_name. For example, euler in the first and dpmpp_2m in the second.

6. Show the second result

   Add a VAE Decode node and a Preview Image node. The output of KSampler goes into the samples input of VAE Decode, the VAE output of Load Checkpoint into its vae input, and its output into the images input of Preview Image.

   Now one run gives two pictures side by side.

7. Fix the seed and run (reading)

   Under seed each sampler has a choice of what happens after a run. Set it to fixed in both, or ComfyUI changes the seed after every run and the comparison breaks.

   Save the graph, run it in ComfyUI and look at the two pictures: sharpness, colour, small detail, how faithful to the prompt. Then try a third sampler in the second node, and so on.

8. Then the scheduler (reading)

   Now give both samplers the same sampler_name and change only scheduler in the second.

   • normal and simple: even steps;
   • karras and exponential: more steps at the end, on fine detail;
   • sgm_uniform: the one SD 3.5 is used with;
   • beta: more steps at the start and the end.

   With 20 steps or more the difference is often small; with few steps it is large.

9. Where to start for each family (reading)

   Pairs from ComfyUI's own templates, a good start:

   • SD 1.5 and SDXL: dpmpp_2m with karras, 20 to 30 steps; Pony and Illustrious: euler_ancestral;
   • Flux.1, Flux.2, Qwen-Image: euler with simple;
   • SD 3.5: euler with sgm_uniform;
   • Z-Image: res_multistep with simple;
   • turbo and distilled models: exactly what their template says, some of them take lcm.

   Samplers with "ancestral" or "sde" in the name add fresh noise at each step: the picture does not settle as steps grow, and is more varied.

10. Done (reading)

Save the graph with "File ▸ Save workflow for ComfyUI…", open the file in ComfyUI and run it there. This app runs nothing in ComfyUI itself.

Keep the pair that looks best for your model and delete the second branch, or keep it for the next comparison. The same way compares two cfg values or two step counts: everything the same, one setting changed.

#What the lesson checks

A reading step is done when you press Got it, next. A doing step ticks itself off when the canvas holds:

• Step 2, A picture graph is needed:
  • a Load Checkpoint node.
  • a CLIP Text Encode (Prompt) node.
  • a CLIP Text Encode (Prompt) node.
  • a Empty Latent Image node.
  • a KSampler node.
  • a VAE Decode node.
  • a Save Image or Preview Image node.
  • CLIP of Load Checkpoint into clip of CLIP Text Encode (Prompt) (prompt) (a Load LoRA or CLIP Set Last Layer in between is accepted).
  • CLIP of Load Checkpoint into clip of CLIP Text Encode (Prompt) (negative prompt) (a Load LoRA or CLIP Set Last Layer in between is accepted).
  • MODEL of Load Checkpoint into model of KSampler (first sampler) (a Load LoRA or LoraLoaderModelOnly or ModelSamplingSD3 in between is accepted).
  • CONDITIONING of CLIP Text Encode (Prompt) (prompt) into positive of KSampler (first sampler).
  • CONDITIONING of CLIP Text Encode (Prompt) (negative prompt) into negative of KSampler (first sampler).
  • LATENT of Empty Latent Image into latent_image of KSampler (first sampler).
  • LATENT of KSampler (first sampler) into samples of VAE Decode (first result).
  • VAE of Load Checkpoint into vae of VAE Decode (first result).
  • IMAGE of VAE Decode (first result) into images of Save Image or Preview Image.
• Step 3, Add a second sampler:
  • a KSampler node.
  • MODEL of Load Checkpoint into model of KSampler (second sampler) (a Load LoRA or LoraLoaderModelOnly or ModelSamplingSD3 in between is accepted).
  • CONDITIONING of CLIP Text Encode (Prompt) (prompt) into positive of KSampler (second sampler).
  • CONDITIONING of CLIP Text Encode (Prompt) (negative prompt) into negative of KSampler (second sampler).
  • LATENT of Empty Latent Image into latent_image of KSampler (second sampler).
• Step 4, Make everything the same:
  • seed of KSampler (second sampler) equals seed of KSampler (first sampler).
  • steps of KSampler (second sampler) equals steps of KSampler (first sampler).
  • cfg of KSampler (second sampler) equals cfg of KSampler (first sampler).
  • scheduler of KSampler (second sampler) equals scheduler of KSampler (first sampler).
• Step 5, Change only the sampler:
  • sampler_name of KSampler (second sampler) is not the same file as sampler_name of KSampler (first sampler).
• Step 6, Show the second result:
  • a VAE Decode node.
  • a Preview Image or Save Image node.
  • LATENT of KSampler (second sampler) into samples of VAE Decode (second result).
  • VAE of Load Checkpoint into vae of VAE Decode (second result).
  • IMAGE of VAE Decode (second result) into images of Preview Image or Save Image.

#See also

• Sampler and sampler names
• Scheduler
• Steps
• How lessons work