Starlight Precise 2.6 vs 2.5: What Changed?

Starlight Precise 2.6 vs 2.5: The Verified Changes at a Glance

Starlight Precise 2.6 is a rework built directly on 2.5, keeping the same model intent and hardware requirements while adding a three-level sharpness control, an updated model for better temporal consistency, and up to 15% local speed improvements for AMD and Nvidia GPUs.

Topaz Labs keeps the lineage simple in its documentation:

Whe two versions are similar in model intent and requirements, and 2.6 is literally built upon 2.5.

That makes the Starlight Precise 2.6 vs 2.5 question unusually clean.

The deltas are narrow, documented, and easy to check.

The replacement mechanics matter as much as the deltas.

Topaz Starlight Precise 2.6 replaces 2.5 from release 1.7.0, and docs state 2.5 is essentially accessible via the default third sharpness level.

Here's where it gets tricky:

The Astra model page lists 2.6 as "Previously Starlight Precise 2.5" and credits a key flicker bug fix.

The documentation separately describes an updated model for better temporal consistency.

Whether those two statements describe one change is undocumented.

Treat them as two official claims, not one confirmed fix.

Everything else in the table stayed identical on paper.

Same intent, same requirements, same output ceiling.

Starlight Precise Sharpness Control: How the Three Levels Work

Starlight Precise 2.6 adds a three-level sharpness slider: level 3 is the default and equates to 2.5 output, level 1 gives a lighter, more natural look, and level 2 balances the two. Differences stay conservative, taking the edge off while keeping the model's essence intact.

The Starlight Precise sharpness slider is the one genuinely new creative control in 2.6.

Nothing else in the update changes what you can dial in during a render.

The official framing is restraint. The docs describe conservative differences between levels, intended to take the edge off while keeping the model's essence intact.

So the slider is a taste adjustment, not a second model.

What Each Level Changes

Each level maps to a documented look, and level 3 carries the default status.

Watch this clip to see Starlight Sharp running inside Topaz Video. Third Move Studios covers the sliders and a full 4K export.

Level 3 is the reference point. If you skip the slider entirely, you are rendering the 2.5 look on the updated 2.6 model.

Where Softer Levels Help AI Footage

The documented fit is AI-generated footage with solid structure and composition that still reads soft, plastic, or slightly artificial.

One forum report from the release thread notes that level 1 shows much less image noise and fewer skin artifacts, suggesting it as a possible default.

That is a single user report, not verified behavior.

The honest caveat cuts the other way too. Because docs describe the differences as conservative, over-processing risk is not a documented flaw.

It is simply what to watch on skin and fine texture as you move from level 1 toward level 3.

Temporal Consistency in 2.6: Official Fixes and Reported Caveats

Official Topaz material credits Starlight Precise 2.6 with an updated model for better temporal consistency, and the Astra model page separately credits a key flicker bug fix, while early user reports mostly agree consistency improved but include unresolved threads about stutter and performance.

Temporal stability is where official claims and forum threads overlap without matching.

Here is the split, labeled by origin.

Confirmed from official material:

• Docs for AMD and Nvidia list an updated model for better temporal consistency.

• The Astra model page credits 2.6 with resolving a key flicker bug.

• Third-party API listings for 2.5 already cited full temporal consistency across frames, so flicker reduction is a 2.5 heritage, not a new 2.6 promise.

The catch:

Whether the flicker fix and the temporal consistency update are one change or two is undocumented. Treat them as two separate official statements.

Reported by users:

• One user found the sharpest setting slightly softer than 1.6.1 output on a difficult test clip, while consistency was far better.

• That single-clip result sits in tension with the docs statement that level 3 equates to 2.5.

• Threads tagged investigating report frame stuttering from 2.6 output and 2.6 performing worse than 2.5, with no official resolution documented.

One user's result on one clip is a reason to test, not a verdict.

Starlight Precise 2.6 Local Performance: Reading the 15% Claim

Topaz documents up to 15% local speed improvements for AMD and Nvidia GPUs in 2.6, with hardware requirements unchanged from 2.5, while diffusion processing still runs significantly slower than core enhancement models and some user reports remain unresolved.

The speed claim needs careful reading before it reaches your render queue.

"Up to 15%" is a ceiling, not a promise. No benchmark methodology or per-GPU breakdown is documented in official material.

The better move:

Read the Starlight Precise 2.6 local claim as possible gains on supported hardware, and plan around the diffusion baseline instead.

Nvidia targets compute capability 8.9, with some success reported on 8.6 cards but no official support. Local rendering runs on all licenses through Neuroserver.

Starlight models process significantly slower than core enhancement models regardless of hardware.

Processing times range from several minutes to multiple days, depending on video length, resolution, frame rate, and system.

Context: 2.5 reached local desktop rendering in Topaz Video 1.4.0 for Nvidia, then expanded to AMD and Mac in 1.6.0, with Mac builds slightly modified and possibly slower.

Same Requirements, New Reports

Hardware requirements are identical for 2.5 and 2.6.

Investigating-tagged user threads include:

• Starlight Precise 2.6 spilling into shared system memory

• 0 FPS on an RTX 5080 16GB during processing

Both reports remain unresolved, with no documented official fix. Treat them as system-specific user reports, not confirmed defects.

Test a short clip on your own hardware before committing to long jobs.

Topaz Video 1.7 Migration: Presets, Projects, and the 2.5 Question

From release 1.7.0, Starlight Precise 2.6 replaces 2.5 in Topaz Video, and the 2.5 look stays essentially accessible through the default third sharpness level. Documented guidance on existing presets, projects, and queued renders after updating is missing, so verify before you update.

Updating to Topaz Video 1.7 swaps the model out from under your workflow.

The replacement is documented; the migration is not.

How saved 2.5 presets, open projects, and queued renders resolve after the update is not documented in available material.

That leaves the risk on you. Treat these as practical checks, not official steps:

• Verify how your saved 2.5 presets resolve once the app lands on 2.6.

• Clear or confirm queued renders before updating, since queue behavior is undocumented.

• Export a short test clip and compare it against a known-good 2.5 render.

Release 1.7.0 also ships Starlight Mini improvements, about 2x faster on Nvidia cards with more than 12GB VRAM and now available to all users.

It adds a Final Cut Pro plugin plus Neuroserver model downloads through Model Manager and the Export Queue.

Availability also depends on the surface you render through.

The Topaz desktop app moved to 2.6, while third-party cloud and API surfaces such as Runware and Satura still list Starlight Precise 2.5.

Which AI-Generated Footage Actually Benefits From 2.6

Starlight Precise 2.6 mainly benefits AI-generated footage with solid structure that reads soft, plastic, or slightly artificial, and the gains come from trying levels 1 and 2. Footage already approved at the 2.5-equivalent level 3 has little reason to re-render.

The docs position Starlight Precise 2.6 for AI-generated video with strong composition and structure that appears soft, plastic, or slightly artificial.

Its focus is realistic faces, skin textures, material, texture, and text.

Here's how that maps to upgrade value:

• Strong-structure AI output that looks soft or plastic gains the most; this is the documented target footage.

• Renders already approved at 2.5 gain little unless you are chasing temporal consistency.

• Very low-quality or below-480p material stays out of scope; Starlight Mini is the documented recommendation there.

The documented weak fits are equally specific:

• Shots with heavy motion blur may be rendered as texture.

• Very low or poor-condition media is out of scope, and content below 480p is not intended.

• Interlaced content must be de-interlaced first.

One honest caveat: official guidance on matching levels 1 to 3 to specific footage types stops at the level descriptions.

The mapping above is editorial interpretation of documented behavior.

This walkthrough shows the full pipeline on real AI-generated clips.

Pay attention to how Bozek Media handles sharpening and the Starlight feature.

The decision rule: run short test exports at levels 1 and 2 on your softest-looking shots.

Keep level 3 for anything already approved.

Unchanged Limits: The 4K Cap, No Preview, and Slow Renders

Starlight Precise 2.6 inherits every operational limit of 2.5: output capped at 3840x2160 locally and in the cloud, no working Preview or Frame Preview, standalone use without other filters, and diffusion-level render times. Upgrading changes the output character, not the capability ceiling.

The version jump adds no new operational headroom.

Every hard limit documented for 2.5 applies to 2.6 in full:

• Output options are Minimum, 2x, 3x, and 4x relative to the source.

• Output is capped at 3840x2160 locally and in the cloud.

• Content already at 4K renders at 3840x2160 regardless of the scale factor.

• Preview and Frame Preview will not work; export a few seconds instead.

• The model is standalone and cannot be combined with other filters.

• Square-pixel footage at or under 4K UHD is recommended to avoid aspect ratio shifts.

Local rendering still runs through Neuroserver on all licenses, and diffusion processing keeps its own timeline regardless of the cap you set.

One limit hits AI-generated video specifically: text that is too small, blurry, or inconsistent may be interpreted as texture or pattern.

AI video output often contains exactly that kind of marginal on-screen text.

The practical result: the upgrade question is about look and consistency, not new capability.

Starlight Precise 2.6 Upgrade Questions, Answered