Nodes/DARASK Nodes/DARASK Exif Apply (SDXL / Checkpoint)
ComfyUI Node

DARASK Exif Apply (SDXL / Checkpoint)

The EXIF time-machine for the one-file checkpoint era

By daraskme·Created 4 months ago·Updated 3 months ago· 1
DARASK Exif Apply (SDXL / Checkpoint)
  • model_override
  • clip_override
  • vae_override
  • model
  • clip
  • vae
  • positive
  • negative
  • positive_text
  • negative_text
  • model_name
  • loras_applied
  • seed
  • cfg
  • sampler_name
  • scheduler
  • steps
  • denoise
  • width
  • height
filepath
fallback_ckpt
positive_prefix
positive_suffix
negative_prefix
negative_suffix
lora_strength_multiplier1.00
skip_loras

If your image collection is SD1.5, SDXL, Pony or Illustrious - anything where one .safetensors file holds the model, both CLIP encoders and the VAE together - then you never wanted the three-loader dance. You wanted CheckpointLoaderSimple, one node, one file. DARASK Exif Apply (SDXL / Checkpoint) is the variant of this pack's metadata-rebuild node that's hard-wired to that: it reads the EXIF/PNGinfo from a filepath, loads the single checkpoint it names, stacks every LoRA the metadata mentions, encodes the positive and negative prompts, and hands you a complete, ready-to-sample pipeline. One input in, seventeen outputs out, zero manual reconstruction.

How it works

Same parser engine as the other Exif Apply variants - it understands A1111/Forge/Reforge parameters text, ComfyUI's embedded workflow JSON, and NovelAI comments, and it recovers model, LoRAs, and the KSampler set (seed, steps, cfg, sampler, scheduler, denoise) plus the original resolution. The difference is purely the loader: it always goes through CheckpointLoaderSimple. So this is the right node whenever you know your sources were SDXL-era gens. It's also the most forgiving entry point into the pack, because it needs exactly one model family and one file layout.

The inputs that matter

Just filepath is required. The useful optional ones:

  • fallback_ckpt - what to load when metadata names a file you don't have. Set this once and batch runs survive the occasional missing model.
  • positive_prefix / positive_suffix / negative_prefix / negative_suffix - wrap the recovered prompts, e.g. a fixed quality suffix for your upscale pass.
  • lora_strength_multiplier and skip_loras - global LoRA dials for taming or dropping recovered LoRAs.
  • model_override / clip_override / vae_override - per-slot overrides; unwired slots keep auto-loading.

Outputs: model, clip, vae, both positive and negative CONDITIONING, then positive_text, negative_text, model_name, loras_applied, seed, cfg, sampler_name, scheduler, steps, denoise, width, height.

Install

Same as the rest of the pack: ComfyUI Manager → search DARASK, install, restart. piexif is handled by the pack's requirements. No downloads.

The one trap

If you point this at an image that was made with a UNET stack (Anima, Flux, Wan - three separate files), the loader can't satisfy the single-file expectation, and the node tells you so with an error that literally suggests switching to the Anima variant. That's by design: use the SDXL variant for the one-file era, the Anima variant for the three-file era, and the Auto-detect variant when the folder can't make up its mind. Since all three share identical outputs, switching is a single node replacement - your downstream wires don't move.

CategoryDARASK

Inputs (11)

NameTypeDefaultDescription
filepathSTRING
model_overrideoptMODEL
clip_overrideoptCLIP
vae_overrideoptVAE
fallback_ckptoptSTRING
positive_prefixoptSTRING
positive_suffixoptSTRING
negative_prefixoptSTRING
negative_suffixoptSTRING
lora_strength_multiplieroptFLOAT1.000–4
skip_lorasoptSTRING

Outputs (17)

NameTypeDescription
modelMODEL
clipCLIP
vaeVAE
positiveCONDITIONING
negativeCONDITIONING
positive_textSTRING
negative_textSTRING
model_nameSTRING
loras_appliedSTRING
seedINT
cfgFLOAT
sampler_nameSTRING
schedulerSTRING
stepsINT
denoiseFLOAT
widthINT
heightINT