jlrouzies-fr/DLSS5-Feeder v1.16.0-beta.7

1.16.0-beta.7

on GitHub

Careful of fake copies. The only download for this project is this releases page. Fake websites and lookalike GitHub repositories exist (#88, #115); this project has no executable installer. The standalone CAREFUL_FAKE_MALICIOUS_FEEDER.txt asset explains how to check what you downloaded. SHA-256 of DLSS5-Feeder-1.16.0-beta.7.zip: 0FF4999C1D00FF6B40555CD127EE83E64494E45D41F6CA56510A791292FBEBBF. The genuine zips are scanned on VirusTotal, for instance the 1.16.0-beta.2 zip; look this release's hash up there yourself.

There is a one-command installer. Download Install-DLSS5Feeder.ps1, put it beside the game's executable, and run powershell.exe -ExecutionPolicy Bypass -File .\Install-DLSS5Feeder.ps1 -Prerelease. The standalone AUTOMATIC_INSTALLATION_AVAILABLE.txt asset explains every option.

Does not work with your game? I can't test every game reported in the issues: I don't own most of them, and I don't have the time to try them all. Read this README section. It gives a prompt for an AI coding agent that reads your logs against the source and opens a pull request if it finds a legitimate fix.

If you are on beta.4, beta.5 or beta.6, update: moving the "Work resolution" slider below 100 crashes the game (#123). This build fixes that, and adds three things asked for in the issues since beta.6. Every binary identifies itself as 1.16.0-beta.7 and embeds release commit 2ed6828; check line 1 of dlss5-feed.log before reporting a retest.

32-bit games: install dlss5-feed.addon32 and host64\dlss5-feed-host64.exe from this zip, as always. The IPC protocol is still v10, so beta.6 and beta.7 halves do talk to each other — but the new wheel support and the corrected helper diagnostics need both new.

Fixes

• #123 — the game crashed when the "Work resolution" slider was moved below 100 (D3D11 games, 64-bit and 32-bit). beta.4 added an orange warning under that slider, shown only below 100, and its text was handed to the UI as a printf format string. The text contains 100% the and 100% and: the C runtime read those as conversions — one invalid, one asking for an argument that was never passed — and ended the process, which is the game, on the first frame the warning was drawn. A config that already said work_resolution below 100 crashed the same way as soon as the add-on's page was opened. Fixed in both add-ons, and every other overlay string was checked for the same mistake.
• #118 — the mouse wheel now works on the in-game panel in games that hide their mouse from ReShade (Castlevania: Lords of Shadow 2 is the measured one). beta.6's notes said this could not be done, because a raw-input sink in the add-on would take the game's own mouse input away. The reporter's answer was the right one: put the sink in the helper process, which has no input of its own to lose. The helper registers it only when the add-on asks — the cursor is over the panel and ReShade cannot see the mouse — so in every other game nothing changes and nothing scrolls twice. If something else in the helper process has already registered raw mouse input (a consumer shipping its own wheel workaround), the helper leaves it alone and says so in its log. High-resolution wheels, which report deltas as small as 1, are accumulated into whole notches: ReShade divides each wheel message by 120 as an integer, and would round every one of them to nothing. The contract (DLSS5_FEED_CAST_WHEEL) is in the README next to DLSS5_FEED_CAST. No IPC change.
• #121 — OpenGL games under Proton/Wine: the log now says what is wrong. Wine advertises the GL interop extensions whatever the Linux driver underneath can do with a Windows handle, so the extension gate passes and the D3D12 fence import then fails. This build does not fix that. It names the failing GL call and the GL error, detects Wine, stops blaming a multi-GPU machine in the "Stopped:" reason, and logs a memory-import probe: line saying whether a D3D12 texture imports — the one fact that decides whether a slower CPU-synchronised fallback can be built at all. If you run OpenGL games under Proton, that line is what #121 needs. Found on the way: the existing texture import FAILED ... GL error lines could only ever print 0x0000, because the error queue had already been drained; they print the real error and the failing call now.
• #119 — the helper's DRED dump (Dred=1) mislabelled its own command list. The helper never named its lists, so its per-frame list showed up as (unnamed -- not this helper's: NGX's or the consumer's) — which sent the reading of a real log the wrong way. The lists are named now (the frame list says outright that it carries the helper's copies, NGX's DLSS and a consumer's inline pass), the dump prints the operations after the stuck one — a stuck resource barrier is batched with whatever follows it — and it counts the list's dispatches, barriers and copies so a run that hangs can be compared with one that does not.

All beta.1 to beta.6 fixes are included.

Verified, and not

The two add-ons and the helper build without warnings; the Vulkan layers and the shader are unchanged from beta.6. The helper's self-test passes 300/300 evaluates here with Dred=1 and the named lists. The wheel sink was exercised against the real helper over the pipe with injected wheel input: nothing forwarded before the add-on asks, notches forwarded while it does, nothing after it stops.

Not run here: the wheel in an affected game (it needs a title that hides the mouse from ReShade), and the OpenGL probe under Proton (it needs Linux). The #123 crash was not reproduced in Zenless Zone Zero itself — the cause was read from the source, and matches the report and its log exactly.