SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts

SECourses Upscaler Pro - SeedVR2 - FlashVSR+ - Face Restoration - Gan Upscalers - Queue System - Fully Automated Movie Restoration and Upscale Studio (Patreon)

Published:

2026-05-05 18:00:00

Edited:

2026-10-13 01:15:09

Imported:

Videos

618821156.mp4

Downloads

Content

SECourses Upscaler Pro

Patreon exclusive posts index to find our scripts easily, Patreon scripts updates history to see which updates arrived to which scripts and amazing Patreon special generative scripts list that you can use in any of your task.

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

Please also Star, Watch and Fork our Stable Diffusion & Generative AI GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)

=======

Latest Installer Zip File : SECourses_Premium_Upscaler_Pro_v8_6.zip

SECourses Ultimate Video and Image Upscaler Pro - SeedVR2 - FlashVSR+ - Face Restoration - Gan Upscalers - Queue System - Fully Automated Movie Restoration and Upscale Studio - with so many amazing new features and full VRAM optimizations, can upscale a whole 120 minutes video fully properly with full resume capability

Quick tutorial video 4K : https://youtu.be/bPWsg8DREiM?si=cMz5avk_etPlwlKY

Version 2.0 of the most awaited Pro Upscaler App published

Almost all of the features completed please test all and let me know if any broken or what new features you need

It has so many features

Fully automated installation and model download on Windows, RunPod, SimplePod, Massed Compute, Linux

It fully supports SeedVR2, FlashVSR+, Image Based Upscalers like Gan upscalers etc

Make sure to follow requirements tutorial fully for torch compile for SeedVR2 it is very requirements depended MSVC C++ Visual Studio CL

Watch demo video above I have uploaded and full tutorial hopefully coming soon

Feature List : Feature_List.png

Windows Requirements

Python 3.12.10, FFmpeg, CUDA 13 cuDNN 9.17, Git, Visual Studio Community Edition with Desktop Development with C++ (all checkboxes checked)

Make sure that your NVIDIA driver is updated (min 590+)

If you get any errors follow below video and its source link

https://youtu.be/DrhUHnYfwC0

https://www.patreon.com/posts/windows-requirements-tutorial-written-post

2 September 2026 Update V8.3

Major Video Chunking, Merging and Audio Synchronization Hotfix

We received detailed reports about frame-count errors, failed chunk merges, audio desynchronization and freezes introduced after V8.2.

We took these reports seriously and rebuilt the affected video-processing pipeline instead of applying a narrow workaround.

Fixed

Fatal delta -1 errors on videos with odd frame counts, such as 157 frames

Incorrect frame totals caused by unreliable MP4 metadata

Audio gradually becoming desynchronized on long videos

Duration drift and Direct stream-copy concat failed validation warnings

Freezes, NoneType errors and crashes while processing the final chunk

Failed or incomplete model outputs being accepted as valid chunks

Audio-only or corrupted temporary files reaching the merge stage

Non-zero and negative timestamp problems caused by B-frames and MP4 edit lists

Incorrect FPS override behavior and VFR-to-CFR conversion

RIFE losing frames at chunk boundaries

Incorrect RIFE frame totals for 2x, 4x, 8x and custom target FPS

LTX 2.5 failures leaving incomplete output files behind

Resume mode accepting stale or incompatible chunks

Chunking problems with apostrophes, Unicode characters and special file paths

Invalid AV1 encoder preset handling

SeedVR2 Flash Attention compatibility issues

Changed and Improved

Frame counts are now verified using actually decoded frames instead of container metadata

Video splitting is now frame-accurate

Every processed chunk is validated before merging

Video, audio, FPS, duration, timestamps and codec compatibility are checked automatically

Unsafe stream-copy merges automatically switch to a robust frame-preserving fallback

Audio is synchronized against the final video timeline instead of individual chunk timing

Long-video chunk readiness checks are now significantly faster

Failed model processes now report the real error instead of continuing with broken files

Temporary concat manifests are unique and cleaned automatically

Error reporting is clearer and more actionable

Compatibility Testing

The new pipeline was tested with:

H.264, H.265, VP9, AV1 and ProRes

MP4, MOV, MKV and WebM workflows

Odd and even frame counts

CFR and VFR videos

Long-GOP and B-frame footage

Silent videos

AAC and other audio configurations

Delayed, shorter, longer and multiple audio tracks

High-bit-depth and different chroma formats

Official Wan and LTX-generated sample videos

Real multi-chunk GPU processing was also completed successfully with:

SeedVR2, FlashVSR+, LTX 2.5, SparkVSR, Image-Based GAN models, RTX Super Resolution, RIFE

Just run Windows_Install_and_Update.bat for update

16 August 2026 Update V8.2

LTX 2.5 upscaler 1 LoRA model download bug fixed

There was a rare bug that was causing audio desynch on longer videos

This issue fixed after reported thank you guys for reporting in details

You can read below details

Just run Windows_Install_and_Update.bat for update

14 August 2026 Update V8

This is a very significant update so please make a fresh install

Now you need to have Python 3.12.10+ installed

LTX 2.5 model implemented

Now it will download the models you use when you first time run them, not auto download

Still it will use our enhanced fast, robust and safe downloader

Moreover, the downloader upgraded and now downloads many times faster than before

Full Int8 ConvRot implemented and their high quality converted versions will be auto downloaded when selected

Application interface fully improved

Interface is much more smooth now

Auto Tune for Max Quality - VRAM system improved for all models

Application moved to Torch 2.13, CUDA 13 with latest compiled libraries

More info coming soon hopefully i will update this post

15 May 2026 Update V5.41

New feature added to the Resolution & Scene Split tab after one of our users requested

Now you can scene split any video you want without running any upscaler process

Just run Windows_Install_and_Update.bat to update

8 May 2026 Update V5.4

Auto Tune for max quality bugs fixed for SparkVSR and how it test and works significantly improved

SparkVSR resolution bugs fixed

SparkVSR is able to utilize text prompting thus after a deep scan preset text prompts added

Supports custom text prompt as well

Previously encoded prompts will be saved and reused and it won't reencode again so it will speed up process of upscale

There was a edge case bug with uncommon resolutions with SparkVSR and this bug fixed and the app is made more robust

Auto Upscale First Frame per Chunk system improved and now it will auto upscale first frame of all chunks according to your Temporal Chunk Length before starting upscale process when enabled

Just run installer to update - no new zip file use latest zip file above

5 May 2026 Update V5.1

It turns out all of the open source versions of SparkVSR out there is not suitable to run on consumer GPU due to lacking optimizations

I have generated BF16 weights and replaced our model downloader to download new BF16 versions and it will be directly used from now on

Get latest zip file, overwrite all files, then delete this folder SECourses_Premium_Upscaler_Pro\SparkVSR and run installer again for BF16 weights to be downloaded

BF16 weights are 20 GB and previously 40 GB were used

No we got both extra space and RAM savings and loading speed increment

But this is not all, I have coded a new feature called as Stage Subprocess Isolation

This is enabled by default and this runs all VAE encoding, VAE decoding, inference upscale steps in sub process and thus now we have significantly lesser RAM and VRAM usage

Auto enabled default

Moreover, now we support FP8 Scaled model as well

FP8 scaled will be automatically generated from BF16 weights and saved in SECourses_Premium_Upscaler_Pro\SparkVSR\models\SparkVSR-fp8-scaled when you first time run it and consequent runs will use it

First time generation may take small time

Furthermore, auto tune function significantly improved thus run it again

Also default Temporal Chunk Length (frame count processed at 1 step) set to 65 for lower VRAM usage

But with auto tune you can find maximum frame count according to your GPU now

With all optimizations I have coded, now the SparkVSR upscale uses like 1/3 RAM and VRAM compared to before

I recommend FP8 Scaled since I didn't notice any meaningful quality difference, model loading ultra fast, RAM usage ultra low and VRAM usage lower

4 May 2026 Update V5.0

New famous SparkVSR model added to our application

Official project page : https://sparkvsr.github.io/

It has all the common shared features of our other upscalers like SeedVR2

Auto tune feature also implemented for this model as well

This model works with reference frames otherwise it is too low quality

Therefore, I have implemented a very special system

First frame of every chunk will be automatically upscaled with your selected model and will be used as a reference frame

Since we already have by default scene based chunked processing, it works amazing

I have added all the favorite models you can pick all, default is SeedVR2 to upscale initial frame of each chunk

Reference frame upscaler models

Download latest zip file (v5 or above), overwrite previous files and run installer

If you get any error, delete venv and run installer again

The installers will automatically download new model necessary files into SECourses_Premium_Upscaler_Pro\SparkVSR folder

28 April 2026 Update V4.8

FlashVSR+ batch upscaling of images in given folder was generating unnecessary sub folders and video files

This bug is fixed now

Just run Just run Windows_Install_and_Update.bat to install / update

5 April 2026 Update V4.7

Input video displayed statistics unified for all upscalers

Now it will display Output FPS as well

Also, if output FPS is overriden via FPS Override (0 = source FPS) or via RIFE it will be displayed now

A custom favicon (the icon displayed at browser tabs) added to the app for easier recognize

Just run Windows_Install_and_Update.bat to install / update

30 March 2026 Update V4.6

I have completey upgraded the color fix process that happens after the upscale

Previously you were not able to do high resolution + batch size SeedVR2 upscale due to 32 bit size limitation

Now there is no limit (we moved to batched fix) and it uses much lesser VRAM and much more optimized with no quality difference

Uniform Batch Size is not enabled by default

Now queue system will be based on individual GPU id therefore with multiple GPU setup, you can run multiple upscales at the same time and queue on each GPU individually

Download latest zip file (still v4 zip file didnt change), overwrite all files, and run Windows_Install_and_Update.bat to install / update

25 March 2026 Update V4.4

ETA feature implemented for all upscalers

The app will try to estimate with highest precision when will entire upscale process end

RTX Super Resolution now has its own override to disable automatic scene detection and splitting when needed, with clearer sizing analysis, better progress feedback, and cleaner cancel handling.

Batch processing is much cleaner across FlashVSR+, GAN, RTX Super Resolution, and SeedVR2 thanks to the new Keep Only Output Files option.

After batch runs, temporary chunks, metadata, and extra artifacts can be removed automatically so only the final outputs remain.

FlashVSR+ batch handling is smarter now and better at detecting videos, single images, and frame-sequence folders.

SeedVR2 received the biggest upgrade: safer VRAM auto-tuning, better handling of Torch compile-related settings, cleaner separation between compiled and non-compiled autotune cache results, and a new advanced option to split encode, upscale, and decode into separate subprocesses for cleaner VRAM release on heavy jobs.

SeedVR2 torch compile is now fully working

Especially brings speed up when Block Swapping not used

The bundled SeedVR2 core was also updated with stronger memory cleanup between phases, reduced tiled VAE VRAM spikes, better compile-dynamic handling, more consistent batch behavior, and improved timing/debug visibility.

Now VAE Encode, Upscale, VAE Decode parts are all different sub processes instead of single process

Thus, 0-VRAM leak or RAM leak and maximum performance

Overall app startup is lighter because several heavy backends are now lazy-loaded instead of being imported immediately, which should reduce overhead and avoid unnecessary startup issues.

FlashVSR+ batch image processing naming issue fixed, it will keep same names

RIFE frame interpolation (2x, 4x FPS) now supports batch folder processing accurately

RIFE frame interpolation (2x, 4x FPS) now supports saving as PNG or JPG (with any quality you want) images directly

Just enable Export as PNG/JPG Sequence checkbox

Lots of bug fixes and improvements with frame sequence processing instead of videos

Download latest zip file (still v4 zip file didnt change), overwrite all files, and run Windows_Install_and_Update.bat to install / update

14 March 2026 Update V4.0

We have added freshly published NVIDIA RTX Super Resolution Upscaler

This upscaler is literally super ultra fast and uses minimal amount of VRAM

I feel like best preset is HIGHBITRATE_ULTRA but you can test all

Download latest zip file, overwrite all files, and run Windows_Install_and_Update.bat to install / update

Let me know if any errors occurs or if you encounter any errors etc

Its impact is weaker than SeedVR2 or FlashVSR+

It can upscale even 8x just fine

7 March 2026 Update V3.6

I have pre-calculated VRAM usages for a lot of aspect ratio and resolutions and they are now all auto downloaded

Starting from 6 GB GPUs to 32 GB GPUs all pre-calculated - actually tested

Best quality best VRAM usage

Few more bugs fixed for automatic VRAM optimization buttons for both FlashVSR+ and SeedVR2

No new zip file just use the latest zip file and run Windows_Install_and_Update.bat to update

28 Feburary 2026 Update V3.3

New amazing feature Auto Tune for Max Quality - VRAM Optimized added for both SeedVR2 and FlashvSR+

This button will test all the parameters according to your input video and output target resolution

It will stop with safe 2 GB empty VRAM and set best quality parameters

When you next time use it, it will scan older test logs and automatically set best parameters according to new video output resolution

I have tested and the input resolution not matters but output resolution total pixel matters

I have uploaded so many pre-made test logs so try Auto Tune for Max Quality - VRAM Optimized and it may immediately set best parameters for your GPU

For this system to work, you should not be using more than 1 GB VRAM of your GPU before running the upscaler app

Final chunk merge algorithm improved and should work even if your chunks were inconsistent FPS or other formats due to inaccurate input video

When upscaling long videos at once with SeedVR2 like 5000 frames, no clip / chunks, it had a massive error at color fixing part - using 100 GB RAM / VRAM

This bug fixed

No new zip file just use the latest zip file and run Windows_Install_and_Update.bat to update

Please have minimum version 3.2 on interface (top left) since lots of bugs fixed with v 3.2

SeedVR2

FlashVSR+

28 Feburary 2026 Update V3

New Optimize Parameters button significantly improved

Upload your video, set 2x or 4x and set your target resolution and click

It will try to set best config according to your GPU

It is not 100% complete yet but should work fairly well, please report me cases where it didn't work

Gradio upgraded to 6.8.0 and now significant performance improvements on the interface smoothness is expected

FlashVRS+ naming was adding unnecessary prefix to output file names, this issue fixed

Some SeedVR2 encoding / logging improvements

SeedVR2 is much better than FlashVSR+ for very low resolution upscales such as 256px into 1024

Default selected SeedVR2 model is set to seedvr2_ema_7b_sharp_fp16.safetensors because it is much better for low resolution upscales which SeedVR2 already shines

SeedVR2 default noise level set to 0 from 0.1 since low resolution videos are already noisy

In Direct Video Compare there was bug that prevented accurate FPS - fixed

In Direct Video Compare a new text alignment feature added

You can see amazing demo here : https://youtube.com/shorts/VLIG9viKKNo

Default Frame Chunk Size in FlashVSR+ increased to 241 from 64

Please download latest zip file, overwrite all other files and run Windows_Install_and_Update.bat

26 Feburary 2026 Update V2.91

New feature added to FlashVSR+

Now it will show you how many DiT tiles it will be based on your Tile Size and Tile Overlap

I recommend you to minimize DiT tiles count with maximizing the Tile Size until you get OOM or starts using Shared VRAM

So for example you can first maximize Tile Size for 2 DiT Tiles and monitor your VRAM usage

If uses shared VRAM or out of VRAM error, reduce Tile Size until you reach 3 DiT Tiles and repeat the test again

Thus gain maximum speed and quality because lesser DiT Tiles = Faster Processing better Quality

For example when upscaling into 3840px - 4K, with tile size 960 I was getting 6 times DiT tiling

However, at this tile size, It used shared VRAM - 31 GB VRAM on RTX 5090

Thus I reduced tile size to 640 and still 6 times DiT tiling, thus it will be still same+ speed and using only 24-28 GB VRAM and thus ultra fast

You can also increase Tile Overlap if you get seam artifacts (repeating patterns)

Copy generated output to input field buttons added to SeedVR2, FlashVSR and Image Gan Upscale tabs

Also all 3 tabs now have checkbox that you can automatically copy output to input after upscale completed

When doing batch processing with FlashVSR+, the Gradio was not displaying status, now it will display

Just run Windows_Install_and_Update.bat to update no new zip file

24 Feburary 2026 Update V2.8

New tutorial video published : https://youtu.be/_WT4C78j5-c

New Flash VSR+ options added : Diffusion Controls (Experimental)

CFG and Noise but normally model doesnt support, you can still try to increase more newer details

Direct Video Compare - Generate Animated Slider Comparison Video speed improved

FlashVSR+ processing log system improved a lot and now displays progress in a single line with also showing speed

Some minor bug fixes

Just run Windows_Install_and_Update.bat to update no new zip file

23 Feburary 2026 Update V2.6

New amazing video comparison features added to 🔍 Output & Comparison > Direct Video Compare tab

You can make comparison and slider comparison videos of any 2 input video with custom text

Check all the features they are literally amazing

Just run Windows_Install_and_Update.bat to update no new zip file

22 Feburary 2026 Update V2.5

In some cases the statistics shown in Chunk Stats was not accurate this bug fixed

In some cases Sizing statistics were wrong like changing upscale ratio, min-max resolution, these bugs fixed

When processing displayed information UI/UX improved

In Video Comparison slider, close button was not clearing the uploaded videos for comparison, this issue fixed

Just run Windows_Install_and_Update.bat to update no new zip file

20 Feburary 2026 Update V2.4

New Global Settings tab added

Just run Windows_Install_and_Update.bat to update no new zip file

20 Feburary 2026 Update V2.3

This is a pretty big update

We have 100% changed the FlashVSR+ backend to a new repo and I have significantly upgraded this repo

The new FlashVSR+ works amazing and I think it is better than SeedVR2 for high res videos upscale like upscaling 720p into higher resolution

Top menu navigation bar updated into a better version and view

FlashVSR+ tab remade and all the features are now working

For lower VRAM a button is added which you can use if you get OOM

Read the updated UI to understand how to use

FlashVSR+ now can upscale images very well as well

Image Based GAN upscalers tab also improved and some bugs fixed

Output & Comparison tab Video Output was not working properly and this issue fix fixed

In Output & Comparison tab, new multi video and multi image comparison sliders added which is super useful to quickly compare multiple videos and images

Lots of various bug fixes made

App is getting closer to be perfect please heavily test it and let me know errors and what features you request

This update was mostly about improving the FlashVSR+ since it is a very fast and amazing video upscaler model

In SeedVR2 tab, Max Resolution setup was not properly working fixed

In FlashVSR+ tab, Full Model + tiled VAE producing noise, now tiled VAE will be auto disabled if full model pipeline selected

In FlashVSR+ tab, use FlashVSR Upscale Factor as upscale

The model supports only 2x and 4x unlike SeedVR2 which supports ratio based upscale

In 🎬 SeedVR2 tab, you can increase Latent Noise Scale to add more new details like increase to 0.3

Image Based - Gan upscale now can upscale videos perfectly fine and Batch Size (Frames per Iteration) is now working to speed up upscaling videos

H265 output and H.265 Tune feature added

Tune like grain or psnr or ssim, etc.

For updating, get the latest zip file, extract and overwrite all files and run Windows_Run_SECourses_Upscaler_Pro.bat file

Also please delete presets folder and first try default and verify it works and then try and save your presets

Check below new screenshots

Fully quality screenshots posted here : https://huggingface.co/blog/MonsterMMORPG/secourses-upscaler-pro

18 Feburary 2026 Update V1.9

FlashVSR+ bug fixes and lots of improvements

Now should work perfect and it is really the highest quality and fastest one imo

FlashVSR+ upscale now shows what is happening, the status and progress both on CMD and also on Gradio Logs panel

Some default parameters tuned to better so you can delete presets folder and remake presets

Hopefully there will be even more tuning

In Output & Comparison tab, Video Output settings were not being saved and loaded with the preset

This issue fixed

Moreover, they settings were not being used, for example when you had set h265, you were still getting h264, this issue also fixed

Fixing this issue literally took 1 day full day coding

Image-Based (GAN) upscaled videos were having frozen frames this issue fixed

Batch size in Image-Based (GAN) were not working and this issue fixed

Chunked previews on Image-Based (GAN) not working and this issue fixed

Model downloader seperated into 2 parts

Default Windows_Download_Models_and_or_Resume.bat will download original BF16 models

Windows_Download_GGUF_and_FP8_Models.bat will download FP8 and GGUF Q8 models

Now we support SeedVR2 GGUF models as well

Download latest v1.9 zip file, overwrite and run Windows_Install_and_Update.bat to update

Run Windows_Download_GGUF_and_FP8_Models.bat if you want GGUF models

Please intensively test and let me know errors and your future requests hopefully I will add all

16 Feburary 2026 Update V1.8

Now you can select GPU to run from very top menu

FlashVSR+ tab remade and now working amazing

Interface greatly improved and supports both 1.1 and 1.0 versions

Health check now checks NVIDIA driver as well, upgrade your NVIDIA driver to latest with clean install

Some global settings were hidden saved and this system removed and every single setting is now saved and loaded with your preset globally

Download latest v1.8 zip file, overwrite and run Windows_Install_and_Update.bat to update

It will download new FlashVSR+ models too

14 Feburary 2026 Update V1.7

A lot of user interface (UI) and user experience (UX) updates and improvements

Global preset system improved and unified

Now every tab and every settings should be saved and loaded with global preset system

Just run Windows_Run_SECourses_Upscaler_Pro.bat to update

12 Feburary 2026 Update V1.6

Folder resume capability added

When you give output folder path e.g. G:\SECourses_Premium_Upscaler_Pro_V1\SECourses_Premium_Upscaler_Pro\outputs\0017 and click upscale, it will resume from the last processed chunk and continue processing and finish the task

Let me know if you encounter any errors

Gradio displayed logs and cmd logs were not synched in some cases

This is due to we can't update Gradio interface too fast or it gets too laggy

You have to wait for full synch to see final output video on Gradio but it will be already saved in folder

Even if you close your browser or Gradio shows error, it will still work

This synching issue improved

Please get latest zip file and overwrite previous files since update logic was missing

And then just run Windows_Run_SECourses_Upscaler_Pro.bat to update