The Loop

Week 2 · SLOP8760

Phase 1 of 4 · The Rig

This week adds

The generation loop: roll one prompt more than once, choose among the rolls, and log the seed that made the one you kept.

Rolling is not re-prompting

One prompt file, five rolls, and the only thing moving between them is the seed: 66740, 66741, 66742, 66743, 66744. The still on this slide is 66743. A seed is the address of the noise a sampler starts from, so the same seed on the same settings returns to the same result.

Source: ComfyUI docs, KSampler — https://docs.comfy.org/built-in-nodes/KSampler

Running the same prompt more than once is called rolling, and it is not a sign the first attempt failed — it’s the default way this loop is used. A generator is a sampler, not an oracle: the same prompt with a different random draw produces a genuinely different result, and the skill this course credits is choosing well among several rolls, not writing a prompt so precise it only needs one. Selection is the actual job. Generation is just what makes selection possible.

The KSampler docs describe seed as “The random seed used for creating the noise”, over the range 0 to 0xffffffffffffffff. That is the whole of it: not a style knob, not a quality setting. Change one parameter and keep the seed, and a specific variable is being tested in isolation; change the seed and a different roll has been generated regardless of what else stayed the same.

The four dials the loop turns

Every generator in this course exposes some version of these, under some interface. Read the published defaults before changing one of them.

Input	Default	Published note
`seed`	—	0 to 0xffffffffffffffff
`steps`	20	—
`cfg`	8.0	“6-8 recommended”
`denoise`	1.0	lower values maintain the input’s structure

Source: ComfyUI docs, KSampler — https://docs.comfy.org/built-in-nodes/KSampler

high-quality samples within only 15 to 20 steps for guided sampling
Lu et al., DPM-Solver++, 2022

Source: DPM-Solver++ — https://arxiv.org/abs/2211.01095

The names on the two lists

Sampler and scheduler are separate choices and they compose. Samplers include euler, euler_ancestral, heun, dpmpp_2m, dpmpp_2m_sde; schedulers include simple, karras, sgm_uniform, exponential, beta. Both lists are read from source and both drift.

Source: ComfyUI source, as of September 2026 — https://github.com/comfyanonymous/ComfyUI

The production log

A roll without a log is a lucky accident the moment the tab closes. Name the graph, the seed, and every parameter that could change the output.

Recorded the loop ladder, tier t1: seed 66740 — Ladder t1 · seed 66740

The ladder

Five rolls of one prompt. Nothing changes between them but the draw.

Roll one — seed 66740
Roll two — seed 66741
Roll three — seed 66742
Roll four — seed 66743
Roll five — seed 66744

Before class

Install the workflow engine if it isn’t already running from last week, and load the graph this course supplies for the semester’s opening weeks.

This week’s exercise

Roll one prompt five times on the supplied graph, log the seed and parameters for every roll, and write a pass or fail verdict for each against the hands, physics, text, and framing checklist. Bring all five rolls, their logs, and your five verdicts to Wednesday’s Dailies.

Reading

High-Resolution Image Synthesis with Latent Diffusion Models
Rombach et al.2022
in the latent space of powerful pretrained autoencoders
Denoising Diffusion Implicit Models
Song, Meng & Ermon2020
10× to 50× faster in terms of wall-clock time compared to DDPMs
DPM-Solver++
Lu et al.2022
high-quality samples within only 15 to 20 steps for guided sampling
ComfyUI: KSampler
Comfy Org
The random seed used for creating the noise

The Loop#

Week 2 · SLOP8760#

This week adds#

Rolling is not re-prompting#

The four dials the loop turns#

The names on the two lists#

The production log#

The ladder#

Before class#

This week’s exercise#

Reading#