deforum / deforum/deforum-stable-diffusion

Feature Request: Use ESRGAN Models from the models/ESRGAN Folder, rather than Existing Models

Open
#308 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
2.3k
Forks
412
PR merge metrics
No merged PRs in 30d

Description

Deforum's Upscaling feature is quite useful. I just wrote a lengthy post on Reddit about it, because someone was asking for a Linux FOSS replacement for Topaz Video AI. With the upscaler and the interpolation features in the Output tab, you basically have the same thing.

The only major limiting factor are the 3-4 ESRGAN models you get to pick from. They aren't very good, as ERSGAN models go, these days, leaving artifacts in details but there's currently no way to add more. Different content often requires a specific model trained on the subject, so currently the only way to do it, is to upscale from the 'Extras' tab, using 'Batch from Directory'

Please consider changing the Upscaler function, so that it uses the models/ESRGAN folder for it's models, rather than building a few into Deforum. That way, there will be no need to include any models and people will be able to use specific models, such as Valar_4x, Nickelback_4x and Lollypop, which you can download here: https://upscale.wiki/wiki/Model_Database

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.