Portrait detail
2× · SeedVR2

A straightforward starting point for larger images. No ComfyUI workflow to install, no Prompt to guess.
Processing requires purchased credits. No free trials or signup credits.


Sample input is AI-generated; the result was processed by the named FAL model.
The same image, on both sides. Drag the handle to inspect what changed.






Original inputs were created with an image-generation tool, then downsampled into test uploads. Every enhanced image was returned by the named FAL model. These are controlled synthetic examples, not customer photos or guarantees of results.
Two real models. Different strengths, different prices. Compare the same input before you choose.
A lower-cost option for general upscaling. Start here for textures, products and everyday images, then inspect the result.
1 credits per output MPOutput: JPG / PNG / WEBPUse this modelExplicit creativity control for reconstructed detail, including portraits. Higher creativity is not automatically more faithful.
7 credits per output MPOutput: JPG / PNGUse this modelThe site limit is our resource cap, not a claim about the model’s absolute maximum. Parameters and prices were checked against the FAL API on November 4, 2026.
| Parameter | SeedVR2 | Crystal |
|---|---|---|
| Available here | 1×, 2×, 4×; 4K long-edge mode | 1×, 2×, 4×; 4K long-edge mode |
| Default | 2×; original proportions retained | 2×; original proportions retained |
| Creativity | Not supported | 0–10 |
| Noise scale | 0–1 | Not supported |
| Text Prompt | Not accepted by either endpoint | Not accepted by either endpoint |
| Output format | JPG, PNG, WebP | JPG, PNG |
| Site output limit | 64 MP per image | 64 MP per image |
| Official maximum resolution | Not specified in the published schema | Not specified in the published schema |
| Upload | JPG, PNG, WebP · 10 MB maximum | JPG, PNG, WebP · 10 MB maximum |
| FAL reference cost | US$0.001 / MP | US$0.016 / MP |
Choose 2× and begin with noise scale 0.1. Keep the seed at 42 while comparing settings so random variation does not obscure the test.
Try 0, 0.1 and 0.2 on the same image. Judge edges and texture, not sharpness alone; higher noise is not necessarily more accurate.
The online API exposes a focused set of controls. A local ComfyUI workflow can expose different precision, tiling, memory and preprocessing choices, and needs its own GPU setup. Results are not guaranteed to match.
One credit is $0.01. Your chosen model and output size determine the cost. No subscription, no signup bonus, no hidden “free” trial.
1 credits per output MP
7 credits per output MP
Billed in whole output megapixels, rounded up, with a 1 MP minimum. 2× width and height means 4× the pixels; 4× means 16×.
Failed processing returns your credits. Higher creativity does not cost extra within the same model and output size.
For a focused project
$5.00500 credits · one-time paymentKeep a balance ready
$15.001,500 credits · one-time paymentFor larger image batches
$30.003,000 credits · one-time paymentPurchased credits do not expire. All packs use the same credit value: no bonus-credit dilution. Taxes, if applicable, are calculated at checkout. Checkout opens after payment setup.
No. All new image processing uses purchased credits. You can inspect the already-generated examples without uploading or spending credits.
Their underlying processing costs differ. SeedVR2 is 1 credit and Crystal is 7 credits per output MP here. The displayed quote is calculated from your selected model and resulting pixel dimensions.
For the same image and model, 4× produces four times the output pixels of 2×. Whole-MP billing and the 1 MP minimum can change the ratio for very small images.
No. These Crystal and SeedVR2 endpoints accept an image and numeric controls, not a text Prompt. We only send parameters the selected endpoint supports.
The image is sent to FAL for processing and stored privately by this service. Our uploads expire after 7 days and outputs after 30 days. FAL has its own processing and retention policies; read the privacy notice before uploading sensitive images.
Not necessarily. AI can reconstruct plausible detail, including incorrect detail. Use the comparison to inspect faces, text and repeating patterns; the output is not documentary proof.