It's a workflow to upscale image several times, gradually changing scale and parameters.
Versions
Short version uses a special node from Impact pack.
Older versions are not better or worse, but they are long and expanded. They provide more control, allowing to tune intermediate steps. They requre WAS Node Suite.
Installation and dependencies
Install Impact pack custom nodes (should be a newest possible version, as for 12.02.2024);
(optional) Install SD Prompt Reader custom node;
Download and open this workflow;
Recommended resources
Complete list of all resources chosen in nodes across versions:
VAE: vae-ft-mse-840000-ema-pruned. Standard, recommended for more details.
Realistic:
SD models: epiCRealism. How to substitute: with any realistic model.
Analog Madness. How to substitute: with any photorealistic model.
Upscaler model: 4xNMKD Superscale. How to substitute: download 4xESRGAN (or take it from Automatic1111).
Textual Inversions: bad_prompt_version2. Lower the weight for unusual compositions.
Deep Negative. Use 16T for base generation, and 2T for upscale.
bad-Hands-5. Use with 0.8 weight.
How to substitute Textual Inversions: just skip.
Anime:
SD models: CamelliaMix. How to substitute: with any anime model.
OnlyAnime. How to substitute: with any anime model.
Upscale Model: 4xNMKD YandereNeo XL. How to substitute: download 4xUltraSharp.
Textual Inversions: EasyNegative. Recommended.
badhandv4. Only use if hands are not important. Otherwise use bad-Hands-5 (but it might change the whole picture).
How to substitute Textual Inversions: just skip.
Description
Now faster! I ask the author of Impact Pack for a related update and he implements a new node! Such wow. Open development, baby. 11.02.2024
FAQ
Comments (7)
Interested in helping me with a workflow ,
High quality realistic images
Wild cards / multi line random text in prompt box , per a line
Auto lora trigger words added to prompt
Mupitle lora generation for different images set each set own batch and own lora In use
Change clothing option
Controlnet and able to cycle Poses
Cycle prompts
constant clothing and face
It is possible to Add your two workflow to or anything else you have mind
you said "Connect this to "Iterative Upscale (Image)" node to generate image instead of loading one."
what is 'this node' ?
i dont get it
VAE Decode (near the note) → Iterative Upscale (below the note)
@Postpos Prompt outputs failed validation IterativeImageUpscale: - Required input is missing: pixels
this is what i got..
Well, I can't be more straightforward than error text already is.
I have same issue. Connected output from 'VAE Decode' above to 'pixels' in the 'Iterative Upscale (image)' item but still get error despite being connected.
It seems that some of the nodes are on bypass, you need to right click the node and bypass to toggle it back on or reload