DVP Tools / Image Enhancement

Professional Mastering with Upscale.AI

Bridge the resolution gap. Upscale AI video and images to 4K with SeedVR2 and Topaz, take the frame rate to 60fps with Topaz Interpolate, and finish Core Workflow 2.0 packets with delivery receipts and manifest export.

Start upscaling Quick start Pipeline Methodology

Quick start

Master your footage in four moves

Upscale.AI is a finishing console. It sends your files to upscaling and frame-rate models on fal.ai, then records what was delivered.

Ingest and Target

Upload your AI-generated JPG, PNG, or MP4. Select your Upscale Factor (2× or 4×) and target model based on your source generator.

Deploy the Pipeline

For video, set the frame rate on the Enhance tab. The 'Pipeline' tab upscales first and then retimes, so you get both in one run.

Review and Compare

Use the live Before/After slider to inspect texture restoration. Download your production-ready 4K / 60fps master once processing completes.

Export the Delivery Record

When a Core Workflow packet is active, download the Delivery Manifest so the final output URL, preset, source, scale, fps, and filename travel with the production packet.

Methodology

The Enhancement Stack

We use a multi-stage enhancement process to ensure that upscaled footage feels organic and detailed, rather than "digitally sharpened."

The Models

Each one does a different job. Which one runs depends on the option you pick.

SeedVR2 Standard image upscaling. Stays close to the source.
Topaz Precision / Starlight Video upscaling. Precision stays close to the source; Starlight rebuilds detail on soft footage.
Topaz Interpolate Builds the in-between frames for 48 or 60fps.
Topaz Bloom 2 Creative image upscaling. Invents new detail.
Face Restore Topaz Precision with face enhancement on.
Real-ESRGAN The Anime image option, and the Draft video engine.

fal.ai Integration

Processing happens on high-end cloud GPUs via fal.ai. This allows you to upscale 4K video in minutes rather than hours on a local machine.

Tip

A fal.ai API key is required. Processing is billed per job, typically costing $0.01 per image and $0.10+ per video clip.

Delivery Console

When Upscale.AI receives a Core Workflow 2.0 production packet, it recommends a finishing preset, applies matching controls, and keeps delivery receipts for finished outputs.

Apply Finish Preset sets source, scale, model, and fps.
Delivery Manifest exports packet-linked output metadata as JSON.
Pipeline Resume can recover pending long-running pipeline jobs when worker state is still available.

Controls

Mastering the Stack

Source Selection

Match the upscaler to your generator. Select "Kling" for cinematic realism or "Anime" for illustrated styles to ensure the correct texture bias.

Super-Resolution (2× / 4×)

2× suits social content. 4× takes a 1080p frame to 8K, so most 4K deliveries only need 2× from a 1080p source. On video, anything over 4096 px comes back as H.265, which most browsers will not preview. The download is fine.

Frame Interpolation

Takes 24 or 30fps footage to 48 or 60fps by building the in-between frames. It does not change the frame size.

Creative (Bloom 2)

Generative upscaling. It adds new texture instead of only sharpening what is there, so check faces and fabric against the original.

Batch Processing

Upload a folder of stills from a shoot. Upscale.AI processes them sequentially, saving results to your session history automatically.

Batch Queue

The Batch Queue tab is the run itself: watch each job's state as it works through the list, and use Clear Queue to abandon what has not started. Queue a night's worth of stills and walk away.

Quick Presets

Match the source instead of guessing a factor. Standard (photos), Portrait/Face, Runway Gen-3/4 and Topaz Max Quality each set the stack for material that behaves differently: AI-generated frames need different handling from a real photograph.

Download Comparison

⊞ Download Comparison writes the before and after out as one image, which is what you actually send a client or attach to a note. The slider proves it on screen; this proves it in their inbox.

Finishing Presets

Use Clean 4K, Cinematic 60, Face Safe, Animation Preserve, Social Delivery, or Balanced to set the tool for a clear delivery target before spending fal.ai credits.

Before/After Slider

A critical QC tool. Drag the handle to inspect skin textures, fabric weaves, and atmospheric bloom at 100% zoom before downloading.

Professional Workflow

Enhancing for Delivery

Upscale then Interpolate

Do resolution first, then frame rate. The Pipeline tab runs them in that order.

The History Tab

Your last 25 results are stored in local browser history. Download links still expire after fal.ai purges output files, so save important masters and manifests right away.

Production Packet Delivery

If you arrived from DVP Vault, use the delivery target panel at the top of the tool. Apply the recommended preset, run the job, then download the manifest for the final packet record.

Creative: Subtle vs Creative

Use 'Subtle' for character close-ups to keep the likeness. Use 'Creative' for wide landscapes where added detail is less likely to change who or what is in the shot.

Cost Estimation

Check the price badge before hitting 'Run Pipeline'. 4K video enhancement is compute-intensive and carries a higher fal.ai credit cost.

Long Clips on the Draft Engine: Cut, Upscale, Rejoin

Precision and Starlight do not use this. fal lists five minutes as the longest source for the Topaz video models, so cut anything longer yourself. The Draft engine (Real-ESRGAN) takes 1000 frames at a time, and a clip too long for that gets a Cut plan. Copy the ffmpeg command it gives you to split the source at those boundaries — a stream copy, so no re-encode and no generation loss. Upscale each part, then use Rejoin: copy concat.txt and the join command to stitch the finished parts back into one file. Keep the parts in order; the concat list is what defines the sequence.