SeedVR2 and FlashVSR+ Studio Level Image and Video Upscaler Pro Released

[FurkanGozukara]

SeedVR2 and FlashVSR+ Studio Level Image and Video Upscaler Pro Released

Full tutorial: https://www.youtube.com/watch?v=bPWsg8DREiM

It has been long waited to have a studio level video and image upscaler app. Today we have publishing the version 1.0 of SECourses Ultimate Video and Image Upscaler Pro. It is supporting SeedVR2, FlashVSR+, Gan based upscalers, RIFE frame interpolation, full queue system, full batch folder processing, scene / chunked based processing and many more. It is fully working on every cloud and consumer GPUs like RTX 2000, 3000, 4000, 5000 series and H100, H200, B200, RTX PRO 6000. We are installing app with latest Torch and CUDA versions atm all fully automatic with pre-compiled libraries. Even Torch compile is fully and automatically working.

📂 Resources & Links:

💻 SECourses Ultimate Video and Image Upscaler Pro Download Link : [ https://www.patreon.com/posts/Upscaler-Studio-Pro-150202809 ]

🚆 Requirements Tutorial : https://youtu.be/DrhUHnYfwC0

🛠️ Requirements Written Post : [ https://www.patreon.com/posts/Windows-AI-Requirements-Setup-Guide-111553210 ]

👋 SECourses Discord Channel for 7/24 Support: [ https://bit.ly/SECoursesDiscord ]

ℹ️ SimplePod Register : https://simplepod.ai/ref?user=secourses

ℹ️ SimplePod Template : https://dash.simplepod.ai/account/explore/100/ref-secourses/

ℹ️ RunPod Register : https://get.runpod.io/955rkuppqv4h

ℹ️ RunPod Template : https://get.runpod.io/SECourses_CU13

ℹ️ Massed Compute Register : https://bit.ly/SECoursesMassedCompute

⏱️ TIMESTAMPS

00:00:00 Announcing SECourses Ultimate Video & Image Upscaler Pro V1

00:00:22 Ultra-Fast High-Quality Image Upscaling Demo (1024 to 4096px)

00:00:40 Low GPU Usage & Prompt-Free Workflow Explained

00:01:00 Single Step Upscaling & Advanced Optimization (Torch Compile)

00:01:20 Introduction to Video Upscaling & Custom Live Comparison Tool

00:01:45 Video Quality Comparison vs Topaz Labs & Open Source Models

00:02:08 Cloud GPU Support (RunPod, Massed Compute) & Local Execution

00:02:20 Dynamic UI Features: Real-Time Input Analysis & Settings

00:02:59 Scene-Based Upscaling Workflow with PySceneDetect Integration

00:03:18 Advantages of Scene-Based Processing & Resume Functionality

00:03:48 Visualizing Processed Scene Chunks & Adjusting Split Settings

00:04:09 Automating Full-Length Movie Upscaling (120+ Minutes)

00:04:26 Analyzing Scene Detection Statistics & Frame Count Distribution

00:04:51 RIFE FPS Interpolation Support & Run Logs Interface

00:05:10 Visual Demonstration of Upscaled Video Details & Future Models

00:05:30 Extensive GPU Compatibility (RTX 2000-5000, H200, B200, CUDA 13)

00:05:52 VRAM Error Handling, Health Check & Recovery Tools

00:06:19 Comprehensive Queue System & Changing Parameters Live

00:06:48 Canceling Processes & Error Recovery Without App Restart

00:07:11 Output Folder Structure, Subfolders & Saving Processed Data

00:07:56 Debugging with Executed Command Logs & Run Summaries

00:08:18 One-Click Windows Installation, Auto-Updates & Model Hash Check

00:08:38 Critical Windows Requirements for Torch Compile & Python 3.10

00:08:57 Installation Instructions for RunPod & Massed Compute Cloud

00:09:21 Detailed Feature List: Face Restore, GANs, FlashVSR+ & More

00:09:55 System Diagnostics, Reliability, Safety & UX Features

00:10:28 Batch Processing Support for Image & Video Folders

Core features: Multi-tab, multi-pipeline app (SeedVR2 / GAN / RIFE / FlashVSR+ / Face Restore) with unified resolution strategy + global face restoration controls.

SeedVR2: Video/image/folder inputs, first-frame preview, smart model defaults/constraints, and advanced VRAM/performance options (offload/BlockSwap/VAE tiling/compile/attention backend/cache toggles).

GAN / RIFE / FlashVSR+: GAN upscaling for images/videos/folders w/ metadata-based model detection; RIFE interpolation (model, FPS modes, scale/precision, static-skip, recursion); FlashVSR+ diffusion upscaling (version/mode/scale + VRAM-aware tiling/unload options).

Output & comparison: MP4/PNG/sequence outputs, full ffmpeg encoding controls (codec/CRF/preset/pixfmt), audio copy/re-encode/remove, 2-pass, FPS override, global RIFE post-process (incl. chunk-safe), and rich comparison tools (slider/SxS/overlay + HTML slider + auto comparison videos + “compare any 2 videos”).

Long runs & batching: PySceneDetect chunking + fallback, lossless vs fast splitting, resume partial runs, per-chunk cleanup, batch processing across pipelines, and an app-wide FIFO queue with snapshot-isolated settings.

Presets & reliability: Universal presets + per-tab presets, migration/guardrails, persistent global settings/profiles, safe cancel/cleanup, OOM detection + guidance, partial output salvage, collision-safe naming, telemetry/command logs, and validation/fallbacks (ffmpeg/audio utilities).

Diagnostics & UX: Health checks (ffmpeg/CUDA/tools/disk), GPU-safe detection, model registries, env var integration, file upload/path entry, previews, auto input detection, estimators, quick links, progress/ETA, CUDA cache clear, and Gradio --share support.

Video Transcription

• 00:00:00 Today I am announcing and publishing the version 1 of the SECourses Ultimate Video and Image Upscaler Pro.

• 00:00:07 This is the beginning of the studio-level image and video upscaler application, video restoration application.

• 00:00:15 There will be so many other new features and models. Currently, we are supporting SeedVR2 and FlashVSR+.

• 00:00:22 So let me show you some of the upscales. This is an image upscaling. At image upscaling, it is amazing, ultra-fast and ultra-high quality.

• 00:00:30 So from 1024 to 4096 pixels. The left is the original, the right is the upscaled.

• 00:00:39 So you see from this image into this image. This is next-level upscale. There are no prompting. There isn't anything. You just upload your image and upscale.

• 00:00:53 And this works on very low GPUs as well because image upscale is very lightweight.

• 00:00:56 Here, another example. Again, 1024 to 4096 pixels.

• 00:01:00 So you see how much details it adds. This is a single step upscale. This is ultra-fast. We are supporting all kind of optimizations,

• 00:01:10 even including block-to-swap or torch compile.

• 00:01:15 So you can run the torch compile as well. We are supporting all kind of features with the SeedVR2.

• 00:01:20 So let's see some of the video upscales.

• 00:01:29 This is a custom implementation made by me. There is no Gradio video upscale comparison, but we have it in our application, so you can use this to make a live comparison of the upscaled video.

• 00:01:45 Here another example, let's watch this one as well.

• 00:01:54 As you can see, it is working amazing. The upscale level is the best one currently as an open source. This upscale is actually even better than the paid Topaz Lab upscale.

• 00:02:08 Because this upscale can fully utilize your GPU. You can also run this upscaler on cloud machines like RunPod, like Massed Compute, like SimplePod, so you are not limited to your own GPU.

• 00:02:19 But it is fully working locally as well. So one another feature is that when you upload a video, it will show you all of its features,

• 00:02:30 like input image, upscale target, actual preprocessing, output, output resolution, how many FPS, how many total frames.

• 00:02:40 And as you change any settings, it will get updated immediately. Like this, or like this. Or you can set bigger resolution like this.

• 00:02:47 Moreover, we are fully supporting the scene-based upscale. We are using the PySceneDetect. Therefore, when you upload a multiple scene video,

• 00:02:59 it scans and show all the scenes. Let me demonstrate you.

• 00:03:04 So now, as soon as the upload of the video completed, it will start analyzing input. So it is analyzing resolution, scene, how many frames it has. As you are seeing the status.

• 00:03:18 What this does is that you are upscaling each scene individually. Therefore, you don't need to use bigger batch size,

• 00:03:30 bigger than the scene itself. And the scenes are very consistent. And let's say you cancel it for any reason or something happened,

• 00:03:39 you can continue from the left scene. So this scene-based upscale is extremely important. It makes it lighter weight.

• 00:03:48 You see the scene results as they appear, as they are processed. Let me show you. So you see there are 2 chunks. Scenes actually.

• 00:03:59 So it will split your video into scenes based on the PyScene. You can change the PyScene settings from here as well if you are not satisfied with its splitting.

• 00:04:09 This application is designed with the mindset of a ultimate upscaler studio. Therefore, you will be able to fully automatically upscale even full movies, like 120 minutes movie,

• 00:04:22 you will put it here and you will upscale it fully automatically.

• 00:04:26 So the scene detection has been completed. This video has 233 scenes. Therefore, it will be split into 233 scenes,

• 00:04:39 and each scene will be automatically processed. You can see that average scene duration is 169 frames, minimum is 61, and maximum is 960.

• 00:04:51 So this is extremely useful. Moreover, we support RIFE FPS as well. If you don't see full of the menu, just click here and you will see other tabs as well.

• 00:05:01 So during the processing, it will show you all of the run logs here. It will show you the process here. Let's watch another video.

• 00:05:10 (Silence)

• 00:05:17 As you can see, it is working pretty amazing, adding a lot of details. And hopefully if there are better models, if there are better upscaling, I will add them.

• 00:05:30 Currently we are supporting all these SeedVR models. Soon I will implement NF4 as well. This application is working on every GPU including RTX 5000 series or B200 or H200.

• 00:05:43 So we are supporting both cloud GPUs and consumer GPUs like RTX 4000, 3000, 2000 series, 5000 series. We are supporting every GPU out there with CUDA 13 support.

• 00:05:52 Moreover, if you get out of VRAM error, it will show you how to improve your VRAM usage. Moreover, we have health check,

• 00:06:00 which you can run health check and see all of the features, what is supported.

• 00:06:07 Furthermore, since we are running with processing mode, even if you get out of VRAM error, you don't need to restart application. You can just change the settings and click upscale again.

• 00:06:19 We also support full queue system. So let's upscale this. It will start upscaling. And let's change some parameters. Let's go to output comparison,

• 00:06:29 and enable global, enable RIFE from here. Then let's return back. When you enable RIFE 2x FPS, it will show you the output FPS here.

• 00:06:37 And when I click upscale, it will be added into the queue. When you go to the queue here, you will see the queued item here. You can cancel the queue from here by delete selected.

• 00:06:48 We also support full cancel of the process. So confirm cancel and cancel. And it will cancel the running process.

• 00:06:59 And it should start the next queue. You can cancel next queue as well and start upscaling again. So without restarting the application, you can fix your error, you can add multiple queues.

• 00:07:11 This application will get hopefully even better. Just use it and give me feedback. Moreover, all of the upscales will be saved inside outputs folder,

• 00:07:22 with subfolders, with their processes, executed commands. If you are enabling 2x FPS, it will also have 2x FPS as well.

• 00:07:30 So everything will be fully saved. Even the processed chunks will be saved. Input chunks will be saved. So every information will be saved that you will be able to see like run summary.

• 00:07:44 Or we will also save the executed commands inside your installation folder. Let me show you that. So you will see the executed commands as well, like the SeedVR2 executed commands.

• 00:07:56 These will be very, very useful to find what is the issue or what parameters were used. So this application is being developed with full professional way of programming.

• 00:08:18 How to install detailed tutorial will be made later. But for installation, just run Windows_Install_and_Update.bat file. That's it. This will be used for both installation and updating.

• 00:08:31 It will also fully automatically download all of the necessary models with hash verification and with 16 connection speed.

• 00:08:38 Make sure to follow Windows requirements if it doesn't work. Especially torch compile is extremely requirements dependent.

• 00:08:47 If you don't use torch compile, it should work with just Python if you have just Python 3.10. Otherwise, you need to follow these requirements. This is important. Don't forget that.

• 00:08:57 For RunPod and Massed Compute, just follow the Massed Compute instructions and RunPod SimplePod Premium Upscaler instructions txt files. Use the templates and instructions given in those files, as usual.

• 00:09:12 The link to download this application will be in the description of the video. So please read this link as usual with full details.

• 00:09:21 Moreover, I recommend you to download this feature list and read it. We have so many features as you can see. We support SeedVR2,

• 00:09:30 GAN-based upscalers, output comparison, RIFE FPS increase, face restoration, FlashVSR+. This is not fully tested yet. It will be hopefully soon.

• 00:09:42 We have system diagnostics and environment control, UX operational convenience. We have scale throughput and long run processing, reliability, safety and recovery.

• 00:09:55 So we support all the features that is useful as an image and video upscaler studio application. This is professional level. Moreover, if you have other requests,

• 00:10:07 just message me from Patreon, from YouTube, from Discord, and I will try to add them. I want to make this application the very best both image and video upscaler out there.

• 00:10:15 With all of the models out there, with all of the features out there, fully automated. It can run even hours or days and upscale your given video or images.

• 00:10:28 We also fully support batch processing at every upscale. So you can do all kind of batch processing as well. And it will upscale all the videos or the images inside given folder.

• 00:10:41 So use it, give me feedback, and hopefully more detailed tutorial will be soon.