Fix SD crash on repeated generation, improve material pipeline UX#201
Fix SD crash on repeated generation, improve material pipeline UX#201fernandotonon merged 1 commit intomasterfrom
Conversation
📝 WalkthroughWalkthroughDeferred Stable Diffusion texture generation was integrated into the LLM-driven material workflow: missing textures detected in LLM output now trigger SD generation and delay final material completion. Img2img editing APIs and UI were removed; QML styling and preview cache-busting were added. Changes
Sequence DiagramsequenceDiagram
participant User
participant QML as MaterialEditor (QML)
participant LLM as LLMManager
participant Core as MaterialEditorQML
participant SD as SDManager
participant Worker as SDWorker
participant Ogre as Ogre Engine
User->>QML: Request material generation
QML->>LLM: Send generation request
LLM->>Core: onLLMGenerationCompleted(script)
Core->>Core: Detect missing texture refs (regex)
alt Missing textures detected
Core->>Core: set m_sdPendingForMaterial = true
Core->>QML: sdPendingForMaterialChanged()
QML->>QML: Show "Generating texture..." + progress
Core->>SD: generateTexture(textureName, prompt)
SD->>Worker: generateTexture()
Worker->>Worker: recreateContext()
Worker->>Ogre: Save generated texture file
Worker->>SD: generationCompleted(filePath)
SD->>Core: onSDGenerationCompleted(filePath)
Core->>Ogre: Apply texture to material
Core->>Core: clear m_sdPendingForMaterial
Core->>QML: aiGenerationCompleted(finalScript)
else No missing textures
Core->>QML: aiGenerationCompleted(script)
end
Estimated Code Review Effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly Related PRs
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
📝 Coding Plan
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: bccc7afbc5
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (!isOgreAvailable()) { | ||
| // Update texture name for preview even without Ogre | ||
| m_textureName = fileName; | ||
| emit textureNameChanged(); | ||
| emit sdTextureGenerated(outputPath); | ||
| return; |
There was a problem hiding this comment.
When onLLMGenerationCompleted() auto-starts SD, the new m_sdPendingForMaterial flow is only cleared at the bottom of onSDGenerationCompleted(). This early return exits before that cleanup, so in any run where isOgreAvailable() is false — or Ogre::Root::getSingletonPtr() hits the similar return a few lines later — aiGenerationCompleted() never fires and sdPendingForMaterial never resets. The result is a stuck "Generating texture..." state and a material that is never applied in non-Ogre/headless setups.
Useful? React with 👍 / 👎.
| m_pendingMaterialScript = cleanedText; | ||
| m_sdPendingForMaterial = true; | ||
| emit sdPendingForMaterialChanged(); | ||
| sdManager->generateTexture(cleanPrompt, 0, 0, texName); |
There was a problem hiding this comment.
This new pending path assumes the auto-triggered SD request will end in onSDGenerationCompleted() or onSDGenerationError(), but MaterialEditorQML still has no generationStopped handler even though the texture panel can call stopTextureGeneration() while sdIsGenerating is true. If a user cancels an LLM-triggered texture run, m_sdPendingForMaterial remains true forever and the deferred aiGenerationCompleted() never happens, leaving the editor blocked on the texture step and dropping the generated script.
Useful? React with 👍 / 👎.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (4)
src/SDWorker.cpp (1)
197-212:⚠️ Potential issue | 🟠 MajorHonor stop requests before entering
generate_image().
m_stopRequestedis reset beforerecreateContext(), but it is not checked again until aftergenerate_image()returns. A stop fired while the context is being rebuilt still launches a full generation, and the UI does not enter the generating state until that rebuild is done. Bail out immediately afterrecreateContext()if stop was requested before callinggenerate_image().Suggested guard
// sd.cpp crashes on second generate_image() call with the same context. // Recreate the context before each generation to ensure clean state. recreateContext(); + if (m_stopRequested.load()) { + m_isGenerating.store(false); + emit generationStopped(); + return; + } if (!m_ctx) {🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/SDWorker.cpp` around lines 197 - 212, After calling recreateContext() inside the generation routine, check m_stopRequested.load() and bail out immediately if true: set m_isGenerating.store(false) and return before emitting generationStarted() or calling generate_image(); optionally emit a suitable stop signal (e.g. generationStopped()) if your codebase expects it. This ensures a stop requested while rebuilding the SD context prevents starting a full generation; locate the check near recreateContext(), m_ctx handling, and before the call to generate_image().src/MaterialEditorQML.cpp (3)
3002-3016:⚠️ Potential issue | 🟠 MajorDon't emit the normal AI completion signal after SD fails.
qml/MaterialEditorWindow.qmltreatsaiGenerationCompletedas the happy path: it validates, applies, and sets a success status. Emitting it unconditionally aftersdGenerationErrorturns a texture-generation failure into a false positive in the main workflow.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/MaterialEditorQML.cpp` around lines 3002 - 3016, In onSDGenerationError(...) remove the unconditional emit of aiGenerationCompleted(m_pendingMaterialScript) so a failed SD does not signal the happy-path completion; instead just clear the pending state: set m_sdPendingForMaterial = false, emit sdPendingForMaterialChanged(), and clear m_pendingMaterialScript without emitting aiGenerationCompleted from MaterialEditorQML::onSDGenerationError; keep the existing emits sdGenerationError(...) and sdIsGeneratingChanged() so the UI can handle the failure path.
2971-2991:⚠️ Potential issue | 🔴 CriticalDon't overwrite
m_materialTextin the deferred SD path.When
m_sdPendingForMaterialis true, this edits the currently loaded Ogre material andupdateMaterialText()serializes that material back intom_materialText. The QML completion handler then callsapplyMaterial()without pushingm_pendingMaterialScriptback first, so the old material can be applied instead of the LLM-generated script.💡 Suggested fix
- // Apply texture to current material pass - Ogre::Pass *pass = getCurrentPass(); - if (pass) { - Ogre::TextureUnitState *texUnit = getCurrentTextureUnit(); - if (!texUnit) { - texUnit = pass->createTextureUnitState(); - updateTextureUnitList(); - if (m_textureUnitList.size() > 0) { - setSelectedTextureUnitIndex(m_textureUnitList.size() - 1); - } - } - if (texUnit) { - texUnit->setTextureName(fileName.toStdString()); - } - } + if (!m_sdPendingForMaterial) { + // Manual SD generation updates the currently edited material immediately. + Ogre::Pass *pass = getCurrentPass(); + if (pass) { + Ogre::TextureUnitState *texUnit = getCurrentTextureUnit(); + if (!texUnit) { + texUnit = pass->createTextureUnitState(); + updateTextureUnitList(); + if (!m_textureUnitList.isEmpty()) { + setSelectedTextureUnitIndex(m_textureUnitList.size() - 1); + } + } + if (texUnit) { + texUnit->setTextureName(fileName.toStdString()); + } + } + } @@ - updateMaterialText(); + if (!m_sdPendingForMaterial) { + updateMaterialText(); + }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/MaterialEditorQML.cpp` around lines 2971 - 2991, The current flow writes the edited Ogre material back into m_materialText via updateMaterialText() even when m_sdPendingForMaterial is set, which can overwrite the pending LLM-generated script; modify the code in the texture-apply/completion path to guard against this by skipping the updateMaterialText() and any assignment to m_materialText when m_sdPendingForMaterial is true (or, alternatively, restore m_pendingMaterialScript into m_materialText immediately before updateMaterialText() if you prefer to serialize the pending script); reference m_sdPendingForMaterial, m_materialText, m_pendingMaterialScript, updateMaterialText(), and applyMaterial() to locate and fix the logic so the deferred SD material is not overwritten.
2793-2827:⚠️ Potential issue | 🟠 MajorGenerate every missing texture before resuming the deferred apply.
This loop stops after the first unresolved
texturereference. With the new normal-map flow, it's easy to produce scripts that reference both diffuse and normal textures, so the later unresolved references still reachaiGenerationCompleted()as broken material inputs.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/MaterialEditorQML.cpp` around lines 2793 - 2827, The loop in MaterialEditorQML.cpp currently stops after the first missing texture (breaking), which leaves other referenced textures unresolved; modify the logic around sdManager->generateTexture so it collects all missing texture names found in cleanedText (while iterating QRegularExpression matches) and triggers generation for each (calling sdManager->generateTexture(cleanPrompt, 0, 0, texName) per missing texture) instead of breaking after the first; keep setting m_pendingMaterialScript = cleanedText and m_sdPendingForMaterial = true (and emit sdPendingForMaterialChanged()) once if at least one texture was scheduled, set sdTriggered = true when any are scheduled, and remove the break so all missing textures are enqueued for generation.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@src/LLMManager.cpp`:
- Around line 242-243: The instruction that tells the model to invent texture
filenames must be made runtime-aware: in LLMManager.cpp where the material
prompt is composed, wrap the texture-generation sentence so it is only included
when sdManager->isModelLoaded() (or under an ENABLE_SD guard), otherwise either
remove it or replace it with "fall back to AVAILABLE TEXTURES" wording;
alternatively move this sentence out of the static prompt and into the runtime
user prompt path used by MaterialEditorQML (which already checks
sdManager->isModelLoaded()), so references to invented files are never emitted
when SD is unavailable (look for code that composes prompts in LLMManager::...
and MaterialEditorQML and update the logic around sdManager->isModelLoaded()).
In `@src/SDWorker.cpp`:
- Around line 161-179: recreateContext() currently sets m_isModelLoaded to false
on failure but does not emit model state signals or return failure, so callers
(e.g. SDManager) never get modelUnloaded/modelLoadError and QML stays stuck;
change recreateContext() to return a boolean status, set
m_isModelLoaded.store(false) on failure, emit the appropriate signal(s)
(modelUnloaded() or modelLoadError(QString) as used by SDManager) immediately
when creation fails, and return false so callers can surface the error before
any generation error is emitted; update callers to check the boolean return and
avoid proceeding with generation when false.
---
Outside diff comments:
In `@src/MaterialEditorQML.cpp`:
- Around line 3002-3016: In onSDGenerationError(...) remove the unconditional
emit of aiGenerationCompleted(m_pendingMaterialScript) so a failed SD does not
signal the happy-path completion; instead just clear the pending state: set
m_sdPendingForMaterial = false, emit sdPendingForMaterialChanged(), and clear
m_pendingMaterialScript without emitting aiGenerationCompleted from
MaterialEditorQML::onSDGenerationError; keep the existing emits
sdGenerationError(...) and sdIsGeneratingChanged() so the UI can handle the
failure path.
- Around line 2971-2991: The current flow writes the edited Ogre material back
into m_materialText via updateMaterialText() even when m_sdPendingForMaterial is
set, which can overwrite the pending LLM-generated script; modify the code in
the texture-apply/completion path to guard against this by skipping the
updateMaterialText() and any assignment to m_materialText when
m_sdPendingForMaterial is true (or, alternatively, restore
m_pendingMaterialScript into m_materialText immediately before
updateMaterialText() if you prefer to serialize the pending script); reference
m_sdPendingForMaterial, m_materialText, m_pendingMaterialScript,
updateMaterialText(), and applyMaterial() to locate and fix the logic so the
deferred SD material is not overwritten.
- Around line 2793-2827: The loop in MaterialEditorQML.cpp currently stops after
the first missing texture (breaking), which leaves other referenced textures
unresolved; modify the logic around sdManager->generateTexture so it collects
all missing texture names found in cleanedText (while iterating
QRegularExpression matches) and triggers generation for each (calling
sdManager->generateTexture(cleanPrompt, 0, 0, texName) per missing texture)
instead of breaking after the first; keep setting m_pendingMaterialScript =
cleanedText and m_sdPendingForMaterial = true (and emit
sdPendingForMaterialChanged()) once if at least one texture was scheduled, set
sdTriggered = true when any are scheduled, and remove the break so all missing
textures are enqueued for generation.
In `@src/SDWorker.cpp`:
- Around line 197-212: After calling recreateContext() inside the generation
routine, check m_stopRequested.load() and bail out immediately if true: set
m_isGenerating.store(false) and return before emitting generationStarted() or
calling generate_image(); optionally emit a suitable stop signal (e.g.
generationStopped()) if your codebase expects it. This ensures a stop requested
while rebuilding the SD context prevents starting a full generation; locate the
check near recreateContext(), m_ctx handling, and before the call to
generate_image().
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 2c2209df-0718-4033-b265-ff137cae4ad2
📒 Files selected for processing (11)
qml/MaterialEditorWindow.qmlqml/PassPropertiesPanel.qmlqml/TexturePropertiesPanel.qmlqml/ThemedTextField.qmlsrc/LLMManager.cppsrc/MaterialEditorQML.cppsrc/MaterialEditorQML.hsrc/SDManager.cppsrc/SDManager.hsrc/SDWorker.cppsrc/SDWorker.h
SD crash fix: - Recreate sd_ctx before each generate_image() call — sd.cpp corrupts internal state after first generation on macOS Metal backend - Remove img2img (generateFromImage/editTexture) entirely — crashes on Metal regardless of context recreation - Remove edit texture UI panel (was using img2img) Material generation pipeline: - Defer material apply until SD finishes generating texture, avoiding intermediate state where material references nonexistent texture - Unified progress bar: shows LLM progress then SD texture progress - Register generated_textures dir as Ogre resource location at startup so textures from previous sessions are found immediately - LLM system prompt now encourages texture_unit for realistic materials (was "Do NOT add texture_unit") Dark theme fixes: - Add themed background/label to all inner GroupBoxes in PassProperties and TextureProperties panels - Add themed contentItem to all CheckBoxes in PassPropertiesPanel - Fix ThemedTextField placeholder color to use disabledTextColor - Fix Information/Preview/Texture Selection GroupBox labels Other: - PBR template → Normal Map template (only normal maps supported) - Updated AI prompt placeholder text - Preview cache-busting for regenerated textures with same filename Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
fernandotonon
force-pushed
the
fix/sd-second-generation-crash
branch
from
bccc7af to
54a76c3
Compare
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (3)
src/SDWorker.cpp (1)
205-213: Context recreation on every generation is expensive but necessary for Metal stability.The workaround is well-documented. However, note that there's no debouncing in the call chain (
MaterialEditorQML::generateTextureFromPrompt→SDManager::generateTexture→SDWorker::generateTexture), so rapid successive calls will each trigger a full context free/reallocate cycle.Consider adding a guard or debounce mechanism at the
SDManagerlevel if users can trigger rapid generation requests via QML. This is a performance optimization rather than a correctness issue.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/SDWorker.cpp` around lines 205 - 213, Rapid successive calls to MaterialEditorQML::generateTextureFromPrompt → SDManager::generateTexture → SDWorker::generateTexture will each recreate the expensive Metal context; add a debounce/guard in SDManager::generateTexture to coalesce or reject rapid requests (e.g., an atomic "isPending" flag or a short QTimer debounce window) so that repeated calls within the debounce interval are ignored or merged and only one call forwards to SDWorker::generateTexture; ensure the guard is thread-safe, clears when the generation completes or errors, and preserves existing error/success signaling to the caller.src/MaterialEditorQML.cpp (1)
2794-2840: Only the first missing texture triggers SD generation.The
breakat line 2828 means only the first missing texture in the material script will trigger SD generation. If the LLM generates a material with multiple missing textures (e.g., diffuse + normal map), only one will be auto-generated.This may be intentional to simplify the workflow, but consider documenting this limitation or queueing multiple texture generations if multiple missing textures are common.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/MaterialEditorQML.cpp` around lines 2794 - 2840, The loop over texture matches only triggers SD generation for the first missing texture because of the break; remove the break and instead collect all missing texture names (or immediately call SDManager::generateTexture for each missing texName) while setting m_pendingMaterialScript = cleanedText and m_sdPendingForMaterial = true once before triggering generation(s), set sdTriggered = true if any were queued, and ensure onSDGenerationCompleted handles multiple completions (or tracks remaining jobs) so aiGenerationCompleted is emitted only after all generateTexture calls finish; key symbols: the QRegularExpression loop, texName, sdManager->generateTexture(...), m_pendingMaterialScript, m_sdPendingForMaterial, sdTriggered and onSDGenerationCompleted.qml/PassPropertiesPanel.qml (1)
76-86: Consider extracting a reusable themed checkbox label pattern.The same
contentItem: Textblock is repeated in multiple changed checkboxes. A small reusable component (e.g.,ThemedCheckBox.qml) would reduce duplication and keep future theme changes centralized.Also applies to: 96-106, 116-126, 217-227, 259-269, 301-311, 343-353
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@qml/PassPropertiesPanel.qml` around lines 76 - 86, Duplicate contentItem: Text blocks in multiple checkbox instances should be replaced with a small reusable component (e.g., ThemedCheckBox.qml) that exposes properties for text, textColor, indicatorWidth and spacing and internally implements the shared contentItem behavior (using parent.text, color, leftPadding: indicatorWidth + spacing, verticalAlignment: Text.AlignVCenter). Create ThemedCheckBox.qml and update each checkbox instance that uses the repeated block (the blocks around contentItem: Text referencing parent.text and parent.indicator.width + parent.spacing) to use this component and bind the appropriate properties so theme changes are centralized.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@qml/PassPropertiesPanel.qml`:
- Around line 76-86: Replace usages of the SystemPalette-derived textColor bound
on the label Text elements with the app theme color from MaterialEditorQML (e.g.
change bindings that read "color: textColor" in the Text contentItem blocks to
use "color: MaterialEditorQML.textColor" or the appropriate MaterialEditorQML
label color property); update every matching Text instance inside the checkbox
rows (the contentItem: Text blocks that reference parent.text and leftPadding:
parent.indicator.width + parent.spacing) so labels follow the app theme rather
than the OS palette.
---
Nitpick comments:
In `@qml/PassPropertiesPanel.qml`:
- Around line 76-86: Duplicate contentItem: Text blocks in multiple checkbox
instances should be replaced with a small reusable component (e.g.,
ThemedCheckBox.qml) that exposes properties for text, textColor, indicatorWidth
and spacing and internally implements the shared contentItem behavior (using
parent.text, color, leftPadding: indicatorWidth + spacing, verticalAlignment:
Text.AlignVCenter). Create ThemedCheckBox.qml and update each checkbox instance
that uses the repeated block (the blocks around contentItem: Text referencing
parent.text and parent.indicator.width + parent.spacing) to use this component
and bind the appropriate properties so theme changes are centralized.
In `@src/MaterialEditorQML.cpp`:
- Around line 2794-2840: The loop over texture matches only triggers SD
generation for the first missing texture because of the break; remove the break
and instead collect all missing texture names (or immediately call
SDManager::generateTexture for each missing texName) while setting
m_pendingMaterialScript = cleanedText and m_sdPendingForMaterial = true once
before triggering generation(s), set sdTriggered = true if any were queued, and
ensure onSDGenerationCompleted handles multiple completions (or tracks remaining
jobs) so aiGenerationCompleted is emitted only after all generateTexture calls
finish; key symbols: the QRegularExpression loop, texName,
sdManager->generateTexture(...), m_pendingMaterialScript,
m_sdPendingForMaterial, sdTriggered and onSDGenerationCompleted.
In `@src/SDWorker.cpp`:
- Around line 205-213: Rapid successive calls to
MaterialEditorQML::generateTextureFromPrompt → SDManager::generateTexture →
SDWorker::generateTexture will each recreate the expensive Metal context; add a
debounce/guard in SDManager::generateTexture to coalesce or reject rapid
requests (e.g., an atomic "isPending" flag or a short QTimer debounce window) so
that repeated calls within the debounce interval are ignored or merged and only
one call forwards to SDWorker::generateTexture; ensure the guard is thread-safe,
clears when the generation completes or errors, and preserves existing
error/success signaling to the caller.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 57984d87-fbd1-4215-9885-434846abeb0b
📒 Files selected for processing (11)
qml/MaterialEditorWindow.qmlqml/PassPropertiesPanel.qmlqml/TexturePropertiesPanel.qmlqml/ThemedTextField.qmlsrc/LLMManager.cppsrc/MaterialEditorQML.cppsrc/MaterialEditorQML.hsrc/SDManager.cppsrc/SDManager.hsrc/SDWorker.cppsrc/SDWorker.h
✅ Files skipped from review due to trivial changes (3)
- qml/ThemedTextField.qml
- src/LLMManager.cpp
- qml/TexturePropertiesPanel.qml
🚧 Files skipped from review as they are similar to previous changes (3)
- src/SDManager.h
- src/MaterialEditorQML.h
- qml/MaterialEditorWindow.qml
| contentItem: Text { | ||
|
|
||
| text: parent.text | ||
|
|
||
| color: textColor | ||
|
|
||
| leftPadding: parent.indicator.width + parent.spacing | ||
|
|
||
| verticalAlignment: Text.AlignVCenter | ||
|
|
||
| } |
There was a problem hiding this comment.
Use app theme colors for checkbox labels to avoid palette drift.
On Line 80, Line 100, Line 120, Line 221, Line 263, Line 305, and Line 347, checkbox label text is bound to textColor (SystemPalette), while the rest of the panel largely uses MaterialEditorQML.*. If app theme and OS palette diverge, label contrast can regress in dark mode.
🎨 Suggested fix
contentItem: Text {
text: parent.text
- color: textColor
+ font: parent.font
+ color: parent.enabled
+ ? MaterialEditorQML.textColor
+ : MaterialEditorQML.disabledTextColor
leftPadding: parent.indicator.width + parent.spacing
verticalAlignment: Text.AlignVCenter
}Also applies to: 96-106, 116-126, 217-227, 259-269, 301-311, 343-353
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@qml/PassPropertiesPanel.qml` around lines 76 - 86, Replace usages of the
SystemPalette-derived textColor bound on the label Text elements with the app
theme color from MaterialEditorQML (e.g. change bindings that read "color:
textColor" in the Text contentItem blocks to use "color:
MaterialEditorQML.textColor" or the appropriate MaterialEditorQML label color
property); update every matching Text instance inside the checkbox rows (the
contentItem: Text blocks that reference parent.text and leftPadding:
parent.indicator.width + parent.spacing) so labels follow the app theme rather
than the OS palette.
|
1 participant
Summary
generate_image()corrupts sd.cpp internal state on macOS Metal after first call. Fix: recreatesd_ctxbefore each generation.generated_textures/dir registered as Ogre resource at startup so textures from previous sessions load correctly.texture_unitwith descriptive filenames for realistic materials (was "Do NOT add texture_unit").Test plan
🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Bug Fixes
UI/UX Improvements
Removals