Generation Specs RTX 4070

Sampler tensorrt

Scheduler linear

Steps 1

CFG Scale 1

Seed 42

Target VRAM 12GB

VRAM 12GB - 12GB

GPU Compatibility

Optimal Performance

Runs entirely in VRAM with no memory swapping.

TensorRT 11.4 FP16 BSRGANx2 & LaMa 2K Video Upscale Base: Gemini Video Min 12GB VRAM 1440×2560 RTX 4070 Verified

Dog Wingman Cafe Review: 2K Video Upscale

720p AI video upscaled to native 2K (1440x2560 @ 24fps) on RTX 4070 using local patch LaMa inpainting and TensorRT 11.4 True FP16 BSRGANx2.

Blueprint Summary RTX 4070 Verified

Two-stage video restoration pipeline upscaling 720p Gemini footage to 1440×2560 @ 24fps on an RTX 4070 (12GB VRAM) using patch LaMa watermark inpainting and TensorRT 11.4 True FP16 BSRGANx2.

Pipeline Flow 5 Nodes

Open in Resolver

01 Load Models

UpscaleModelLoader

1 loaders (DiT, CLIP, VAE)

Execution DAG Topology Interactive Visualizer

Drag to pan · Scroll to zoom · Hover wires

Model & Asset Setup Setup Script

Run in your ComfyUI root:

curl -fsSL https://decomfy.com/api/scripts/golden-retriever-tinder-review-2k.sh | bash

Positive Prompt

Two dogs sitting at outdoor sidewalk cafe table in Vietnam with iced drinks, a golden retriever wearing a stylish black beret hat and blue bandana holding up an iPhone showing photo of a cute shiba inu dog, looking over at a rottweiler dog with knowing humorous expression, cinematic film still, detailed fur, vibrant daylight, 8k resolution

Negative Prompt

unclothed, disrobed, lowres, blurry, bad anatomy, deformed hands, extra fingers, low quality, artifacts, watermark, text

Prompt Customization: This workflow is pre-wired and calibrated. Swap this prompt with your own character, subject, or style without breaking the pipeline.

Required Models 0 Models

Required Storage: 0 MB (0 models · No external weights required)

Field Notes RTX 4070 Benchmark

Benchmark: NVIDIA GeForce RTX 4070 (234s execution: 2K TensorRT upscaling & LaMa inpainting only. Base video generated via Gemini/Veo)

1. Processing Scope & Provenance Notice: This recipe is strictly a post-processing restoration and upscaling pipeline on consumer NVIDIA GeForce RTX 4070 (12GB VRAM), NOT an end-to-end local text-to-video generation model. The original base video was generated externally via Google Gemini / Veo. The local RTX 4070 workflow handles two key stages: (1) local patch LaMa inpainting to cleanly erase the bottom-right watermark without temporal flickering, and (2) TensorRT 11.4 True FP16 BSRGANx2 super-resolution upscaling from 720p to native 2K (1440x2560 @ 24fps). 2. Local Patch LaMa Inpainting Architecture: Watermark removal on AI-generated video cannot be treated as a brute-force global inpainting task without causing massive background temporal flicker. We isolate a tight 256x256 bounding box [450:706, 1024:1280] centered over the bottom-right watermark, execute SimpleLama inference on the cropped patch, and paste the inpainted texture back before super-resolution. 3. TensorRT 11.4 True FP16 Super-Resolution: Upscaling 720x1280 up to native 2K (1440x2560) is powered by a serialized TensorRT 11.4 FP16 BSRGANx2 execution engine on an NVIDIA GeForce RTX 4070. By enforcing strict C-contiguous memory layout and direct NVENC streaming, the pipeline maintains ~1.03 fps sustained throughput without VRAM paging. 4. Audio Remuxing and Edge Delivery: Audio tracks are preserved losslessly via two-pass FFmpeg muxing, moving the moov atom to the front (+faststart) for instant zero-buffering HTTP range streaming on Cloudflare Workers edge nodes.

Frequently Asked Questions FAQ

What GPU and VRAM are required to run Dog Wingman Cafe Review: 2K Video Upscale?

This workflow requires a minimum of 12GB VRAM (recommended 12GB VRAM). Tested and verified on NVIDIA GeForce RTX 4070 (12GB VRAM) at 1440x2560 resolution.

How do I resolve missing custom nodes for this workflow?

You can drop the workflow JSON into our client-side Missing Node Auto-Resolver at https://decomfy.com/resolve/ to detect missing nodes and generate install commands, or run the 1-click terminal setup script provided below.

Is this full video generation or a 2K restoration pipeline?

This is strictly a 2K post-processing restoration pipeline on consumer RTX 4070 (12GB VRAM), not a full text-to-video generation model. The base video was generated externally via Google Gemini / Veo; the RTX 4070 executes patch LaMa inpainting to erase the watermark and serialized TensorRT 11.4 True FP16 BSRGANx2 to upscale to 1440x2560 @ 24fps.