Starlight Precise 2.6 sharpness levels vs Sume video upscale tiers

Starlight Precise 2.6 is the Topaz Video model that reworked 2.5 to add three selectable sharpness levels, and Sume Video Upscale 1.0 has nothing equivalent: its only quality control is enhancement_tier (fast, standard, pro), plus a scale ratio. If you need a lighter or sharper look on the same upscale, Sume gives you a tier and a follow-up filter, not a slider.

The Topaz side below comes from the Starlight Precise 2.5/2.6 docs page, read on 2026-10-02. The Sume side comes from the Sume OpenAPI schema for POST /v1/video-upscale-1.0/upscale. The two products are not the same shape: Topaz Video is desktop software with cloud export, Sume is an HTTP API.

What did Starlight Precise 2.6 change?

Topaz's page says 2.5 was reworked into 2.6 so users get more sharpness options. A slider offers three levels: a more natural look with a lighter touch, a balanced middle option, and the sharpest output. The page says the third level equals what 2.5 produced, and that the middle level is the default.

The same page describes the model as aimed at realistic human faces and skin texture, plus materials, texture and text. It recommends inputs at or under 4K, outputs up to 4K, and scale options of minimum, 2x, 3x or 4x. It also warns to expect long render times, and says 2.6 has up to 15% local speed improvements on AMD and Nvidia GPUs.

• Three sharpness levels on a slider (Topaz).
• Output capped at 4K, scale options up to 4x (Topaz).
• Local on Mac, AMD and Nvidia, and cloud export on Mac and Windows (Topaz).

What controls does Sume Video Upscale 1.0 expose?

The Sume request schema is short. It takes a public HTTPS video_url, a scale ratio between 1.1 and 4 (scale_ratio, with upscale_factor as an alias, default 2), an optional duration_seconds between 1 and 30 used for the admit reservation, and enhancement_tier. Provider model ids are selected by Sume and are not accepted in the request.

There is no sharpness field, no face-and-skin mode and no texture toggle. The schema sets additionalProperties: false, so an unknown field is refused rather than ignored. Sume also does not document the model behind the route, so this post does not claim any equivalence with Starlight.

How do the two compare side by side?

The table lists only what each page states.

How do I get a softer or sharper result on Sume?

Pick the tier first, then adjust afterwards. Sume's video filter allowlist includes unsharp and cas, and gblur or smartblur if the upscale came back too crisp. That is a separate paid pass, and the video filter page lists it at $0.02 per encode job with a free /v1/video-filter/check first. The related post on sharpening after an upscale has the filter strings.

If the clip is longer than 30 seconds, cut the hero seconds first with video trim; see the 30-second cap post.

curl -X POST https://api.sume.com/v1/video-upscale-1.0/upscale \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: upscale-demo-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/clip.mp4",
    "scale_ratio": 2,
    "enhancement_tier": "pro",
    "duration_seconds": 8
  }'

Which should I choose?

Choose Topaz Video when you want to audition sharpness levels on your own machine and judge faces and skin by eye. Choose Sume when the upscale is one step in an automated pipeline: submit by URL, poll the job, and keep the output on a Sume-owned URL for the next step.

Sume does not publish a quality comparison against Starlight, and neither page gives one, so test your own footage. A fair test uses the same 5 to 8 second clip at the same scale on both, with a face, some text and a flat gradient in frame, because those are where upscalers differ most. Keep the Sume job's duration_seconds close to the real length so the reservation stays small, and compare the outputs at 100% zoom rather than in a thumbnail. For price, read the live rate from GET /v1/catalog; the earlier per-second cost post shows the arithmetic.