AV ComfyUI Manual
23 / 64
Files
23
BASE GENERATION

The main SDXL pass builds the scene before PEOPLE/PPL

The SDXL stage combines RealVisXL, optional IPAdapter, positive/negative conditioning, two ControlNets and mode-dependent latent/denoise logic.

CONFIRMEDLoader filenames, selectors, sampler controls and links are present in the derived specification
MAIN SDXL · CURRENT MODE 1

The base scene is generated before PEOPLE/PPL

2 → 86/87Model + CLIPRealVisXL · LoRA None
5 / 6Conditioningpositive + negative
417→419→418ControlNetDepth then Canny
535Latent source3 empty / 536 img2img
1KSampler230 + 231 + 600
14VAE Decode→ 565 detail bridge
MODE 1 · CURRENTTXT + CNET2IMGempty latent · batch 4 · denoise 1.0
MODE 2IMG + CNET2IMGVAEEncode 536 · denoise 0.3
01 · MODEL + CONDITIONING

RealVisXL receives LoRA/IPAdapter selection before encoder 5

NodeRoleCurrent value / route
2CheckpointLoaderSimpleRealVisXL_V4.0.safetensors
86CLIPSetLastLayer-1
87LoRA stackAll four slots = None
7 / 168IPAdapterAdvanced / model switch168=1 selects plain SDXL
5Positive encoderString 897 + model CLIP path
6Negative encoderString 814
417→419→418Depth + Canny stackApplied to 5/6
02 · GENERATION MODES

Node 541 changes latent source, base image and denoise together

Compare 602 evaluates 609 against 541. The If node 600 then selects either 1.0 or the user img2img denoise value 0.3. Stored widgets inside 535/592 are subordinate to the linked control.

ModeLatentDenoiseUse
541 = 1 · currentEmptyLatent 3; W/H from 783; batch 4607 = 1.0 through 600TXT+CNET2IMG
541 = 2Selected image 592 → VAEEncode 536608 = 0.3 through 600IMG+CNET2IMG
03 · SAMPLING

KSampler 1 receives parameters from upstream controls, not only its visible widgets

SourceValueDestination
23028 steps · CFG 3.4 · dpmpp_3m_sde_gpu · karrasKSampler 1
231Global seed + randomize policyKSampler 1 and FLUX noise 61
418Positive + negative after ControlNetKSampler 1
535Mode-selected latentKSampler 1
6001.0 or 0.3denoise of KSampler 1
1 → 14Sampled latent → checkpoint VAE decodeDetail stage 565
04 · CHECKPOINT

The first shared visual checkpoint appears after 565/779

CONFIRMED

Topology

Decode 14 feeds detail transfer 565; its output reaches Image Filter 779 and comparer 71.

NOT CONFIRMED

Generated pixels

The JSON does not store actual batch output, memory usage or runtime errors.