[Tracking] Invisible or missing objects, walls and effects #1002
Labels
No labels
accessibility
autorelease: pending
autorelease: tagged
bug
dependencies
documentation
duplicate
enhancement
github_actions
good first issue
help wanted
invalid
question
testing wanted
wontfix
No milestone
No assignees
2 participants
No due date set.
Dependencies
No dependencies set
Reference
MagicBeanDev/Hookshot#1002
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
This issue collects every open report of invisible or missing objects, walls and effects, so there is one place to follow it. The individual reports were closed pointing here; their logs, screenshots and discussion stay in each linked issue.
Summary
Things in the world that are not drawn, or are drawn black or distorted: walls and floors, the ground in the Lost Woods hole, the held torch's flame, cutscene particles, spider webs, the sky.
There are no GitHub reports yet of invisible pushable blocks, doors or enemies. If you see one, please post the place, a screenshot and
hookshot_log.txthere.Affected GPUs and versions
Evidence from the logs
nup.py --verifyon a 23.3 GB.nupfinds all-zero files (CmnTex_Map_Soil_FairyFountain_Floor, Rock_WaterTemple_Gate, GanonStage); a complete dump fixed it for another player.Workarounds
nup.py --verifyon your.nup. If it reports all-zero members, make a fresh, complete dump.Reports merged here (7 issues, about 8 people)
Adding a report
Please update to v2.0.6 first (https://github.com/NotRexed/Hookshot/releases/tag/v2.0.6; Update next to Play in the launcher). Please add: place, screenshot, GPU, version,
hookshot_log.txt, and the size of your.nup.All tracking issues
@NotRexed The spider web on the wall is normal now, but the spider web underground is still black. The exposure issue after defeating Goma still exists. It returns to normal after watching the CG describing the three goddesses, and in the CG at the very beginning of the game, the name on the picture introducing Navi's name has disappeared.
hookshot_log.txt
In the new version, the visual noise in the graphics somewhat affects the viewing experience, and adjusting the sharpness doesn't seem to reduce it much.
Yep, it was a problem with the dump. Getting a complete one fixed the Fairy Fountain problems.
rx9070xt windows 11
hookshot_log.txt
Water Temple, missing walls: still happening on v2.0.9.
Same symptom as #524 and #718. The darkness/black Link issue I had before is fixed on my side since v2.0.4. Only the missing walls remain.
There is a PR for lichii that addresses this, though I haven't been able to test it yet as I had to find a fairy fountain but now I have one and can.
https://github.com/zortos293/Lichii/pull/740
here is a save at a fountain for testing
save.zip
Missing walls (Water Temple, Lost Woods, Castle Town entrance): A player found the cause of missing walls like this: an incomplete .nup dump where 3 textures have a zeroed zstd header (for example Rock_WaterTemple_Gate_M_01_Nrm). The build doesn't notice, and the walls that use those textures aren't drawn. A complete dump (about 23.8 GB) fixes it; dump the game again or copy it again, then rebuild. A builder check for this is being written.
How to identify the incomplete dump behind the Water Temple / Great Fairy Fountain reports. This is my own 23.3 GB
.nup, which reproduces the missing walls:e5b85cea14686c5c6c75d05d57cb78a3116bf0b8f2078361ec1f203ce59417b3a0f8b434a6c3a78cc8eae635b70c8cf8c4871685fa1fb204279169f30efbb62ab9e98be5Three of its romfs files start with zeros:
Tex/Rock_WaterTemple_Gate_M_01_Nrm.bntx.zs,Tex/CmnTex_Map_Soil_FairyFountain_Floor_D_Nrm.bntx.zsandTex/RBake_Scene_GanonStage_G_2_1_d0_t0.bntx.zs. The zeros are in the package's stored data, so nothing in the.nupholds a good copy. #1308 makes the builder put a similar texture in their place and print a warning. Tested in the Water Temple on Linux with an RTX 4070 Ti.Seems the dump 23.3GB is missing stuff. I tried a different dump and fairy fountain came back.
Confirmed: re-dumping fixed it. My old .nup was the incomplete one (the 3 textures started with zeros). After a fresh dump and a rebuild, the Water Temple walls render correctly on my RX 9070 XT. Thanks NotRexed and Joey-Magnus!
Fixed in the v2.0.12 branch, closing when it ships. The builder now stands in for the 3 damaged textures of the incomplete 23.3 GB dump (#1308), so the Water Temple walls and fairy fountains draw. A complete ~23.8 GB dump already works today.
Also in the v2.0.12 branch: the Deku Tree floor web lighting on NVIDIA is restored (#1442), and 40 vertex programs that lowered wrongly (LEA.HI with a constant base) draw again (#1437). Closing the matching reports at release.
I see that zora's fountain have mountain not drawing at the top , we only see trees floating but not the wall. i have seen it at child link.
The zora rivers fallwater effect not working well too
on zora rivers too (and some other places ) i wee some shadows ugly square on the water principally
anb igor tomb is very buggy . i have the 23,8GB dump so all is ok :)
Still happening on v2.0.11.
System: Ryzen 7 5700X3D · RTX 5070 Ti (NVIDIA driver 32.0.16.2114 / 621.14) · 32 GB RAM · Windows 11 Pro build 26300 · Vulkan. 100-180 FPS, performance fine.
Version: v2.0.11 (some screenshots taken on v2.0.10; resumed the same save in the same area on v2.0.11 hours later and the bugs are unchanged).
Progress: ~2+ hours in, young Link, Spanish language. Now on Death Mountain Trail.
Death Mountain Trail: solid black rectangular blocks in the sky (also visible behind the area title card) and bright magenta squares floating in the sky. They look like missing textures or placeholder geometry.
V2.0.12 poe sister problem flame not render .first time you go to this place there is flame on torch but not on hookshot , same like poe sister take flame , they don't have in on there torches
update 2.0.13, missing walls and textures on a lot of zora assets (also zora domain entrance in lost woods) have not tested further.
config:
RX 9070 XT
i7-14700KF
32gb ram
hookshot_log.txt
Same issue here. Doorways and some critical items like the spinning object you can't fully see on top of this moving platform.
AMD Ryzen 7 5800X 8-Core Processor (3.80 GHz)
NVIDIA TITAN X (Pascal) (12 GB)
Windows 11
hookshot_log.txt
nub is 23.8GB
Data point: sharply rectangular flat / untextured wall patch on Linux + RADV (NAVI22), v2.0.13, complete dump verified
Adding this here as the tracking issue asks ("post the place, a screenshot and
hookshot_log.txthere"). I searched the open issues first: #1799 was closed 13 minutes ago as a duplicate of this issue, and #1717 is a different place (Dodongo's Cavern floor), different hardware (RTX 5070 Ti / Windows 11) and disappears after re-entering the game. #993 tracks rainbow / garbled textures on 4–8 GB cards.What happened
A large, sharply rectangular patch of rock wall renders as a flat, almost untextured grey-white slab, with perfectly straight horizontal and vertical boundaries against the correctly textured rock around it. A second spot shows the same abrupt texture change over the upper half of a stone pillar. Geometry and lighting are present — the surface texture is missing or wrong. This is not rainbow/checkerboard noise, not a black surface, and not a missing polygon. Three screenshots are attached.
Where in the game
An interior stone-walled passage with a wall torch (see screenshots). I will follow up with the exact room name on my next run — I did not note the in-game map name when taking the screenshots.
How often
Every time at that spot during one ~20-minute session (2026-11-11 00:39–00:58, UTC+8), seen in three screenshots. All three are from that single run: I have not retested after restarting the game, so I cannot claim it persists across sessions yet.
Version
v2.0.13 (source cfc565a2)— first line ofhookshot_log.txt. Linux native binary, produced by the Builder's update from v2.0.12.I know v2.0.14 is out and I am not on it. Its release notes list only three fixes (build folders with brackets, HDR menu colours, FPS counter size), none texture-related, and the Linux update means a ~29 min rebuild plus shader warm-up. I have not upgraded yet, so I am commenting here rather than ticking the "newest release" checkbox on a new issue form. I will retest on v2.0.14 before opening anything separate.
Graphics card and driver version
PowerColor Radeon RX 6750 GRE 10GB (Navi 22), PCI
1002:73ff/ subsystem148c:2429, 10224 MiB Hynix VRAM, VBIOS113-001-X01.Note:
lspciand the kernel call this "Navi 23 / Radeon RX 6600" because AMD reused the0x73ffdevice ID for the China-only RX 6750 GRE 10GB andpci.idshas no148c:2429subsystem entry. RADV reports it correctly:0x6802006decodes asVK_MAKE_VERSION(26, 2, 6), matching Mesa 26.2.6.mesa 1:26.2.6-1,vulkan-radeon 1:26.2.6-1(RADV); the only ICD isradeon_icd.json, no AMDVLKvulkan-icd-loader 1.4.363.0-1,libdrm 2.4.134-1,linux-firmware* 20261019-17.2.14-arch1-1, moduleamdgpuCPU, RAM and operating system
Intel Core i5-12400F, 16 GB RAM + 16 GB zram swap, Arch Linux x86_64 (rolling), glibc 2.44, niri 26.04-1 (Wayland), SDL3 3.4.18-1, Wayland 1.26.0-1. Monitors: 1920x1080@75 (DP-2) and 1920x1080@60 (HDMI-A-1).
Changes from the defaults, at the time of the screenshots
(The
.iniandhookshot_env.txtfiles themselves are not attached, per CONTRIBUTING; these are the values as text.)Insert-menu rows: Quality preset
Custom; Frame generation2x; Frame pacingOn; Game frame rate cap60; Output resolution2048x1152; Upscalergame's own; Resolution67%; Sharpness30%; Handheld profileOn; HDROff; FullscreenOff; FPS counterOn.hookshot_env.txtlines at that time:NO_MESH,PRECOMPILE=0andNO_FOLLOWwere added earlier to get past a pipeline-compilation hang on this driver; I have no evidence they relate to textures.The game dump is complete — this is not the known 23.3 GB case
The
.nupis25,622,267,860bytes (23.86 GiB), i.e. the complete ~23.8 GB dump, not the truncated 23.3 GB one. I ran the official read-only check:Exit code 0. So the zeroed-zstd-header cause described in this issue does not apply here, and I am also past v2.0.12 where the Builder stands in for those three textures.
Log evidence (
hookshot_log.txtattached)textures 3971 MB (budget 3072 MB)— textures over their budget2674textures retiredeviction events in that one rundevice lost, noover the driver's video-memory allowance; sysfs showed ~6.3–6.6 GB VRAM in use of the driver's ~9.3 GB allowancedisplay: window closed, exiting. This is not a GPU hang (ring gfx timeout) and not an OOM kill.The eviction volume resembles the pattern described in #993, although the visible result here is one flat untextured rectangle rather than rainbow blocks, this is a 10 GB card, and the maintainer marked the 6–8 GB garbled-texture case fixed in v2.0.12. #1774 also lists a small
RX_VK_VRAM_MBas a way to force evictions, so the 3072 MB budget above is a plausible trigger I can A/B.Not tried yet
RX_VK_ASTC_RGBA8=1(the AMD workaround from #1006), AMDVLK instead of RADV, v2.0.14, and a restart with the raised texture budget. On disk I have since changedRX_VK_VRAM_MB=5120,Resolution 100%,Output resolution 1920x1080andHandheld profile Off, but no run has happened since that change (no newerhookshot_log.txt), so I am not claiming those settings fix or fail anything.Other software hooked into the game
No ReShade, no shader-cache tools, no overlays, no frame-generation injectors. Sunshine runs as a system service for streaming but was not streaming during this session and does not hook into the game; the 2x frame generation above is Hookshot's own menu option.
Happy to test a diagnostic build, provide the exact room name / a save position, or attach further logs.
Edit (2026-11-11): attachments, full system details, and a second data point
My first comment above said the screenshots and
hookshot_log.txtwere attached. They were not — my mistake. They are now embedded below. Everything in the original text still stands (same v2.0.13 build, same GPU and driver); this edit adds the missing evidence, the complete system environment, and what I observed in later runs.Attachments
screenshot-1-wall-flat.jpgscreenshot-2-pillar-grey.jpgscreenshot-3-wall-flat-recheck.jpgscreenshot-4-owl-missing-perch.jpgscreenshot-5-ranch-fence-missing.jpgscreenshot-7-ranch-surface.jpghookshot_log.txthookshot_log.txt: hookshot_log.txtscreenshot-6-dark-room.jpgis deliberately not here: it shows a different symptom (a crushed-dark interior) and belongs in the lighting tracker #1004, where I have posted it separately.Full system environment
7.2.14-arch1-1, glibc 2.4426.04-1(Wayland), plasmalogin greeter; SDL33.4.18-1, Wayland1.26.0-11002:73ff, subsystem148c:2429, VBIOS113-001-X01, 10224 MiB Hynix VRAM (sysfsmem_info_vram_total=10720641024bytes)1:26.2.6-1(mesa,vulkan-radeon,lib32-mesa), Vulkan loader1.4.363.0-1,linux-firmware 20261019-1,xf86-video-amdgpu 25.0.0-1; the only ICD isradeon_icd.json(no AMDVLK)AMD Radeon RX 6750 GRE 10GB (RADV NAVI22), Vulkan 1.4,driver 0x6802006=VK_MAKE_VERSION(26,2,6)hdr=0and the log saysvi: QueryDisplayHdrStatus -> SDRVkLayer_LS_frame_generation.json(lsfg-vk-bin 1.0.0-4), MangoHud0.8.4-1, gamescope WSI3.17.0-1, MESA anti-lag, MESA device_selectv2.0.13 (source cfc565a2), Linux native build from the Builder0400486024DEC000v131072,25,622,267,860bytes (23.86 GiB); officialnx2/nup.py --verifyreturnedverified 372587 records, 161114 unique extents: 0 badA note on the GPU name:
lspciand the kernel call this card "Navi 23 / Radeon RX 6600" because AMD reused the0x73ffdevice ID for the China-only RX 6750 GRE 10GB andpci.idshas no148c:2429subsystem entry. RADV reports it correctly. So please read "RX 6600" in my kernel logs as this Navi 22 part, not as a different card.Settings at the time of screenshots 1-3 are in the original comment above. Since then I changed:
Current
hookshot_env.txtlines (as text, per CONTRIBUTING — the files themselves are not attached):NO_MESH,PRECOMPILE=0andNO_FOLLOWare leftovers from working around a pipeline-compilation hang on this driver in v2.0.12. I have no evidence they cause the missing objects. Since #1001 tracks missing terrain on GPUs without mesh shaders, I would appreciate guidance on whether skipping the task/mesh path can remove whole objects; I will A/B removingRX_VK_NO_MESH=1on a later run.More objects entirely absent (later runs)
screenshot-4. I could not recall what the object is, so I am not naming it.screenshot-5.I did not write down the in-game map names for screenshots 1-4, so I cannot name those rooms yet; I can supply them on the next run. Screenshots 5 and 7 are from the ranch area (I believe Lon Lon Ranch, but I have not confirmed the in-game name).
These are not AMD-only. They look like the same family as
other-jared's doorway and "spinning object you can't fully see on top of this moving platform" (NVIDIA TITAN X / Pascal),Djipi71's torch fire not rendering,Caliel666's invisible enemies (RX 9070 XT), andZenkcq's missing Water Temple walls — consistent with the "GPUs: any vendor" note in the issue body.Two symptoms at the same place
Screenshots 5 and 7 were taken at the same ranch spot. In #5 a whole fence section is absent (geometry). In #7 the surfaces are present but flat and almost untextured. Outdoors there the sky, grass, the horse's shading and its cast shadow are all normally lit, so this is not a global lighting or exposure failure — which is why the crushed-dark interior screenshot goes to #1004 and this one does not.
Other variables, stated for completeness
lsfg-vk-bin 1.0.0-4,VkLayer_LS_frame_generation.json) is installed on this system. I have not confirmed whether it is active for the game process. No ReShade or other injectors.2xjuddered badly,Offremoved it,3xlooks smooth,4xjudders again. In the2xrun the log repeatedly printedvk: fsr3fi: 0 of the last 600 made frames by FSR 3.1, 600 without its inputs (the renderer's own made them). Switching to4xalso coincided with a host-memory OOM kill at 02:01:45 UTC+8 (Free swap = 248kB, 16 GiB zram full); there was noring gfx timeout/device wedged, so that was a system OOM, not a GPU hang. I have not A/B'd 3x vs 4x with a log and am not claiming a causal mechanism.Honest gaps in this report
./bin/hookshot 2>&1 | tee hookshot_log-current.txtand add it here.RX_VK_ASTC_RGBA8=1or AMDVLK.Happy to provide a save position, the exact room names, a fresh log from the same run as a screenshot, or test a diagnostic build.
Texture issue throughout Dodongo Cave. AMD9070XT
A new Discord report: the Great Deku Tree with no walls and invisible blocks. If that's you: what's your .nup's size? A complete dump is about 23.8 GB; an incomplete 23.3 GB one hides walls (v2.0.12 and later stand in for its 3 damaged textures, so rebuild on the latest version too).
The water's simulation does not run on its own: waves and ripples stay frozen while the camera is still,
and only update when the camera moves. Walking into the water or standing still does not make it react
either.
V2.0.14
Where in the game: Cow cavern in Death Mountain.
How often: Every time
Graphics card and driver: RTX 5080, driver 621.14
CPU, RAM and OS: AMD Ryzen 7 7700X, 32 GB DDR5-4800, Windows 11 Pro 26H2 (build 26300.9818)
Changes from the defaults: Ultrawide = Monitor's shape (21:9), Frame generation on. Only mod: Xbox UI (romfs, button icons only).
Other software hooked in: No, nothing hooked in
hookshot_log.txt
https://youtu.be/j5WX7dq_zKc
I am also seeing invisible and missing objects in Zora River. Below are screenshots of the same locations in both Hookshot and Fritchii.
Hookshot v2.0.14
Fritchii v5.8
Triage note: damaged dumps produce exactly this symptom family (missing floors/walls) via texture stand-ins — confirmed with a dump whose FairyFountain floor normal was damaged (#1700 was the same cause). Proposed warning-UX improvements so reporters see it before filing: #2237.
Follow-up on v2.1.0: transparent pillars / columns
Two more data points from
v2.1.0(Linux native, Builder build). Same family as the missing-object reports above, but here the geometry is not fully absent — it reads as transparent. A stone pillar/column that should be a solid opaque object shows up as an unlit, hard-edged dark mass that you can see the room through, and where it should join the ceiling or floor the boundary is a clean polygon cut instead of a joint.Both are Inside the Deku Tree; the in-game map corner reads
2Fin the first shot and1Fin the second.screenshot-8-pillar-2f.jpgscreenshot-9-pillar-1f.jpgBoth frames: 2x frame generation active (HUD reads
74 fps / game 37and72 fps / game 36, and the2xbadge is lit), 100% render resolution, 1920x1080 output.Already ruled out
hookshot_version.txtreadsv2.1.0, and the per-build caches were all regenerated by the v2.1.0 Builder today: a full recompile ofhookshot_programs.bin(6,691,159,679 bytes) plus the pipeline/state files at 13:25–13:28, then further rewrites at 22:07 (programs.bin) and 22:36 (pipelines.bin,computes.bin,pipeline_states.bin). The old v2.0.13 set and this fresh v2.1.0 set both show the symptom, which is why I read it as deterministic rather than cache damage.RX_VK_VRAM_MB=5120.Settings at the time of these two screenshots (raw
hookshot_options.inivalues)Current
hookshot_env.txt(as text, per CONTRIBUTING)RX_VK_PRECOMPILEis now1; it was0when I wrote the comment above. On v2.1.0 the first run compiled about 31,637 programs at runtime, with single-frame ASTC uploads up to 1,250 ms, so I went back to startup precompilation.RX_VK_NO_MESH=1is still left over from the v2.0.12 pipeline-compilation hang workaround — I have no evidence it causes this, and I am willing to A/B removing it if you think the task/mesh path could explain partially transparent geometry.Happy to supply whichever is most useful next: a save at the 2F pillar, the same two rooms with
RX_VK_NO_MESH=0, or a run with Hookshot's frame generation fully off to see whether the transparency changes.Update — A/B test pins the transparent geometry to the non-mesh fallback path
Did a clean A/B test on RADV (Mesa 26.2.6, RX 6750 GRE / Navi 22, Hookshot v2.1.0):
mesh_shaders=0+RX_VK_NO_MESH=1(the path my card defaults to): pillars are transparent / do not occlude — this is the bug (same as the screenshots I attached earlier).mesh_shaders=1+RX_VK_NO_MESH=0: the same pillars render correctly.So the invisible/transparent geometry is specifically in Hookshot's non-mesh (task/mesh-disabled) code path, not a cache, VRAM-budget, or global-illumination issue. A full fresh recompile across v2.0.13 → v2.1.0 did not change it, so it is deterministic, not random corruption.
Caveat for the maintainers: on my specific GPU, enabling mesh/task shaders eventually triggered a hard GPU hang (
ring gfx_0.0.0 timeout → MODE1 reset, after several minutes of gameplay, not on entry), so I cannot run mesh-on as a daily workaround. But the correct-vs-broken contrast above still isolates the bug to the fallback path — i.e. the fix should be in the non-mesh geometry/alpha path, not "just tell users to enable mesh shaders" (which is unstable on some RDNA2 cards).Environment: Arch Linux (kernel 7.2.14-arch1-1), niri/Wayland, i5-12400F, 16 GB + 16 GB zram, RX 6750 GRE 10 GB (Navi 22), Mesa/RADV 26.2.6, Hookshot v2.1.0, game dump verified 0 bad.
More reports of this, collected here:
Missing geometry and textures (walls/rocks) in Water Temple - The Legend of Zelda Ocarina of Time
When exploring the Water Temple, the environment geometry (walls, rocks, and structures) fails to render and is completely invisible. The water, the character (Link), and some interactive objects render normally, but the surrounding map structures do not appear, allowing the camera to see through the map.
Graphics Card: NVIDIA GeForce GTX 1080 Ti (11 GB)
Observe the environment to see that the walls, floor, and rock textures are missing/invisible in the Temple of Water.
More reports of this, collected here:
I don't know how imported this is but on NX2Navi i don't have the problem with missing Textures
Tracker upkeep note (merged pull requests and code checked against main
51da0464; no new test run was made for it).Main includes #1308/#2391 for particular damaged-resource forms, #1437 for the constant-base vertex instruction, #1442 for the NVIDIA floor-web lighting case, and several specific compute-fallback repairs. These should not be described as a universal missing-wall fix. The recent transparent-pillar comparison used
RX_VK_NO_MESH=1, which deliberately omits mesh draws; it did not compare native rendering with Compute. A useful next comparison is explicit Compute with that skip unset and a log matching each picture. Count-buffer fallback is independently still unimplemented on51da0464, but no supplied picture has been tied to that operation. Please keep the tracker open and do not treat package size or a limited dump check as full content verification.