VisionDepth3D, also called VD3D, is a 2D-to-3D conversion suite for depth generation, depth blending, stereoscopic rendering, VR output, FPS interpolation, AI upscaling, and live 3D preview.
This updated guide reflects the newer 3D workflow, including the beginner 3D Assistant, the expanded Advanced 3D Controls, the new layered depth behavior, and the new Output & Encoding Presets.
VD3D is designed for two types of users:
- Beginner users can use the 3D Assistant and codec presets without understanding every technical setting.
- Advanced users can open the full stereo controls and tune exact shift, convergence, depth shaping, subject lock, edge repair, and output settings.
-
Prepare your source video
- Check resolution, frame rate, aspect ratio, black bars, and audio.
- If the movie changes aspect ratio, split it into separate sections first.
-
Generate a depth map
- Use the Depth Engine tab.
- Start with a Depth Anything V2 model or Video Depth Anything model.
-
Check depth direction
- Make sure near objects are closer and backgrounds are farther.
- If the result feels inverted, enable Invert Depth.
-
Optionally blend depth maps
- Use Depth Blender when one model gives better subjects and another gives better backgrounds.
-
Load source and depth in 3D Generator
- Load the original video.
- Load the matching depth video.
- Select an output path.
-
Start with the 3D Assistant
- Choose a 3D Style.
- Adjust 3D Strength, Pop-Out, Screen Depth, Depth Comfort, and Subject Stability.
-
Preview before rendering
- Use Red-Blue Anaglyph, HSBS, Shift Heatmap, Overlay Arrows, and convergence guides.
-
Render a short test clip
- Use Clip Range for 10 to 20 seconds before a full render.
-
Choose an encoding preset
- Use Fast Preview for testing.
- Use Balanced Final or High Quality Final for finished renders.
-
Render the final video
| Goal | Start Here |
|---|---|
| Generate a depth map | Depth Engine |
| Blend two depth maps | Depth Blender |
| Convert 2D video to 3D | 3D Generator |
| Use beginner 3D controls | 3D Assistant |
| Fine tune stereo manually | Advanced 3D Controls |
| Move the screen plane | Screen Depth or Screen Plane Offset |
| Create deeper backgrounds | BG Push and Layered Depth Warp |
| Keep subjects stable | Subject Stability, Subject Lock, Subject Plane Lock |
| Reduce edge artifacts | Edge Repair Quality, Edge Masking, Feathering |
| Export VR180 | VR180 Output Settings |
| Improve FPS | FPS / Upscale |
| Upscale video | FPS / Upscale |
| Use live 3D | VD3D Live |
The Depth Engine creates depth maps from images or videos.
A depth map tells VD3D what should appear near, mid, or far when generating stereo 3D.
- Open Depth Engine.
- Select a depth model.
- Choose an output directory.
- Keep default settings for your first test.
- Click Process Image.
- Select your image.
VD3D saves an output such as:
yourfilename_depth.png
- Open Depth Engine.
- Select a depth model.
- Choose an output directory.
- Click Process Video.
- Select your video.
VD3D saves an output such as:
yourvideo_depth.mkv
Controls the internal size used by the depth model.
Lower values:
- faster
- less VRAM
- less detail
Higher values:
- more detail
- more VRAM
- slower
Good starting values:
512x288
704x384
768x432
Controls how many frames process at once.
- Higher batch size can be faster on strong GPUs.
- Lower batch size is safer if VRAM is limited.
If you run out of memory, lower batch size first.
Enable Invert Depth if near and far values are reversed.
VD3D usually expects:
brighter = closer
darker = farther
Depth Normalization stabilizes near and far depth range across frames.
Use it for:
- full movie depth maps
- smoother 3D conversion
- reducing depth flicker
- reducing frame-to-frame breathing
Disable it for:
- fast tests
- maximum speed
- models that are already stable
Enable Save Frames if you want PNG depth frames in addition to a depth video.
Useful for:
- inspection
- debugging
- external editing
- frame-based workflows
The Depth Blender combines two depth sources into one cleaner result.
Use it when:
- one model has better subject edges
- another model has better backgrounds
- depth flickers
- subject edges shimmer
- a single model does not produce enough usable depth separation
Depth Blender can use:
- folders of PNG depth frames
- two depth videos
- Load both depth sources.
- Click Preview Now.
- Scrub frames.
- Adjust blend controls.
- Compare the base depth and blended output.
Controls how strongly the second depth source can add bright near-depth information.
Controls softness between blended regions.
Boosts local depth contrast.
Smooths noisy depth while trying to preserve edges.
The 3D Generator converts a source video and matching depth video into stereoscopic 3D.
It uses:
- source RGB video
- depth video
- 3D Assistant settings
- advanced stereo controls
- subject tracking
- depth shaping
- depth sculpting
- optional depth-order warp
- edge repair
- output and codec settings
The newest 3D updates focus on creating a stronger window into the scene effect, where backgrounds sink inward and subjects stay separated instead of the image looking flat with only foreground pop-out.
The current VD3D shift convention is:
Foreground Shift: usually negative
Midground Shift: usually slightly negative or near zero
Background Shift: usually positive
Recommended range:
| Control | Range |
|---|---|
| Foreground Shift | -5.0 to -10.0 |
| Midground Shift | -0.5 to -3.0 |
| Background Shift | +2.0 to +5.0 |
A good relationship is:
Foreground Shift < Midground Shift < Background Shift
Example:
FG -8.9 / MG -2.5 / BG +2.9
Older presets may not transfer correctly if they used positive foreground and negative background values.
The 3D Assistant is the recommended starting point for most users.
It gives simple controls that map into the advanced render system.
Beginner controls:
- 3D Style
- 3D Strength
- Pop-Out
- Screen Depth
- Depth Comfort
- Subject Stability
The 3D Style dropdown provides preset starting points.
Common styles:
| Style | Use Case |
|---|---|
| Comfortable Cinema | Balanced movie viewing |
| Strong Pop-Out | More forward foreground effect |
| Deep Background | More background recession |
| Close-Up Safe | Safer faces and dialogue scenes |
| VR Comfortable | Lower strain for headset viewing |
| Wide / Deep Scene | Stronger depth for rooms and scenery |
| Clean Edge | Safer edge handling |
| Showcase Mode | Strong demo-style depth |
| Custom | Manual adjusted settings |
When you move a beginner slider manually, the style becomes Custom.
Controls the overall stereo separation.
Higher values:
- stronger 3D
- larger foreground/background separation
- deeper environments when the depth map supports it
- more artifact and comfort risk
Lower values:
- softer 3D
- less strain
- fewer artifacts
Controls how much foreground objects feel like they come toward the viewer.
Higher values:
- stronger foreground presence
- stronger subject pop
- more dramatic 3D
Lower values:
- more natural foreground
- less eye strain
Pop-Out is not the main background-depth control. Use Screen Depth, BG Push, and Layered Depth Warp for deeper environments.
Controls where the neutral screen plane sits.
Meaning:
0 = closer / more pop-out feeling
50 = neutral
100 = deeper behind the screen
Use this when:
- the whole scene feels too close
- the whole scene feels too far back
- the subject needs to sit closer to screen level
- backgrounds need to sit deeper behind the screen
Adjust slowly. Small changes can strongly affect comfort.
Controls safety and strain reduction.
Higher values:
- reduce extreme shift
- improve comfort
- soften aggressive stereo
Lower values:
- allow stronger depth
- allow more pop
- increase artifact risk
Controls how strongly the subject or main depth region is stabilized.
Higher values:
- keeps subjects stable
- reduces subject drift
- helps faces and dialogue scenes
Lower values:
- allows more subject movement
- can create stronger pop
- may be less stable
3D Strength: 50
Pop-Out: 35
Screen Depth: 50
Depth Comfort: 70
Subject Stability: 60
3D Strength: 65
Pop-Out: 35
Screen Depth: 58 to 65
Depth Comfort: 55
Subject Stability: 55
3D Strength: 75
Pop-Out: 55
Screen Depth: 45 to 55
Depth Comfort: 45
Subject Stability: 45
3D Strength: 45 to 55
Pop-Out: 25 to 35
Screen Depth: 48 to 52
Depth Comfort: 80
Subject Stability: 80 to 90
Advanced 3D Controls are for exact stereo tuning.
Use them when:
- a scene is difficult
- the beginner controls are close but not perfect
- you are creating presets
- you need exact control over screen plane, convergence, edge repair, or depth shaping
Controls how strongly near objects are pulled toward the viewer.
Recommended:
Natural: -5.0 to -7.0
Strong: -8.0 to -10.0
Aggressive: -10.0 to -12.0
Controls objects between foreground and background.
Recommended:
Subtle: -0.5
Natural: -0.8 to -2.0
Strong: -2.0 to -3.0
Midground helps prevent the render from looking like only two flat layers.
Controls how far backgrounds sink behind the screen.
Recommended:
Subtle: +1.0 to +2.0
Natural: +2.0 to +3.0
Strong: +3.0 to +5.0
If the background still looks flat, do not only raise Background Shift. Also check BG Push, Depth Pop Gamma, Layered Depth Warp, and depth map quality.
Limits maximum allowed stereo displacement.
Recommended:
Comfort: 0.020 to 0.045
Strong: 0.050 to 0.071
This is a safety clamp. Higher values allow stronger 3D but increase artifact and comfort risk.
Controls overall stereo balance.
Recommended:
Comfort: 0.35 to 0.55
Strong: 0.55 to 0.75
Advanced version of Screen Depth.
This controls where the zero-parallax screen plane sits.
Use it to fine tune the scene after the beginner Screen Depth slider gets close.
Controls how strongly convergence is applied.
Convergence is a placement and comfort tool, not the main depth creator.
Recommended:
0.000 to 0.030 for stable cinematic output
0.030 to 0.120 for stronger tests
Too much convergence can make the whole stereo volume feel like it is sliding around.
Automatically adjusts convergence based on tracked subject depth.
Use it for:
- full movies
- smoother scene transitions
- subject comfort
- fewer sudden depth jumps
Disable it temporarily when testing raw pop-out.
Controls how strongly the tracked subject is pinned near the screen plane.
Higher values:
- more subject stability
- safer faces
- less subject drift
- can reduce strong pop-out
Recommended:
0.25 to 0.85
Locally reduces disparity around the subject depth.
This keeps the subject stable while allowing background depth to remain strong.
Recommended:
0.20 to 0.35
Controls how wide the protected subject depth band is.
Recommended:
0.10 to 0.14
Meaning:
0.08 = tight protection
0.12 = good default
0.18 = wide protection, safer but flatter
If subjects feel unstable, increase width slightly. If too much of the scene feels protected and flat, reduce width.
Controls how depth is reshaped before stereo shift.
Lower values:
- stronger depth punch
- stronger near and mid separation
- stronger background separation when used with BG Push
Higher values:
- softer depth
- more natural scenes
- less cutout effect
Recommended:
0.86 to 0.94 for stronger cinematic depth
1.00 to 1.15 for softer natural depth
Controls where the depth shaping curve focuses.
Recommended:
0.45 to 0.50
Stretch Lo controls near-depth stretch.
Recommended:
0.02 to 0.08
Stretch Hi controls far-depth stretch.
Recommended:
0.94 to 1.00
Boosts foreground contribution.
Recommended:
1.00 to 1.20
Boosts background recession.
Recommended:
1.03 to 1.12
This is one of the most important controls for the new deeper background look.
Adds a subtle rounded shape to near foreground objects.
Use it for:
- faces
- bodies
- large foreground objects
- reducing cardboard-flat subjects
Recommended:
0.03 to 0.08
Too much can look artificial.
Adds sharpening after stereo generation.
Use lightly. Too much can increase halos.
Adds optional depth-based blur.
Use sparingly. Too much can hide depth detail.
The new 3D update improves the illusion that the screen is a window into a scene.
Older behavior could sometimes look like:
flat background
subject popping forward
limited background recession
The new behavior is designed to create:
subjects separated from background
backgrounds sinking inward
stronger room and environment depth
less flat-cardboard feeling
This is mainly done by:
- Cinematic Depth Sculpt
- Depth-Order Forward Warp
Cinematic Depth Sculpt reshapes the depth map before stereo shift.
It protects the tracked subject while pushing far background areas deeper.
It improves:
- background recession
- subject/background separation
- room depth
- wide scene depth
- screen-window illusion
Depth Sculpt is driven by existing controls:
- Depth Pop Gamma
- BG Push
- Subject Plane Lock
- Subject Plane Lock Width
- FG Pop
Depth-Order Forward Warp is an experimental layered warp mode.
Classic VD3D warp is smooth and fast, but it samples the image backward. Depth-order warp adds a forward placement pass where pixels are placed by depth order.
Conceptually:
far pixels are placed first
near pixels are placed last
foreground can sit over background
background can separate more believably
This helps create a stronger layered scene instead of a flat displacement effect.
If exposed in the UI, this controls how much depth-order warp is blended into the classic render.
Suggested meaning:
0 = Classic VD3D warp only
50 = balanced layered warp
100 = strong depth-order warp
Recommended values:
0.35 = safer for faces and motion
0.50 = good default
0.65 = strong room and showcase depth
0.80 = aggressive testing
If background depth improves but subject edges become stressed, lower Layered Depth Warp Strength first.
FG Shift: -8.9
MG Shift: -2.5
BG Shift: +2.9
Max Pixel Shift: 0.071
Parallax Balance: 0.40
Depth Pop Gamma: 0.88
BG Push: 1.08
Subject Lock: 0.65
Subject Plane Lock: 0.25
Subject Plane Lock Width: 0.12
Convergence Strength: 0.015
Dynamic Convergence: On
Screen Plane Offset: 0.000
Floating Window: Off
Edge Repair: High
Layered Depth Warp: 0.50 to 0.65
Use preview before rendering full video.
Good for checking:
- stereo direction
- pop-out
- screen plane
- background recession
- edge artifacts
Shows a half side-by-side preview.
Useful for passive interlaced 3D displays.
Shows stereo displacement as a heatmap.
Use it to check:
- where shift is strongest
- whether foreground and background separate correctly
- whether shift is being clamped
- whether settings are too aggressive
Shows shift strength without direction.
Shows a clipped range to inspect subtle shifts.
Shows shift direction with arrows.
Shows differences between eye views.
Good for spotting:
- excessive disparity
- ghosting
- edge tearing
- unstable depth
Used for inspecting feathering and softened transitions.
Convergence guides add registration marks in preview only.
They help show:
- screen level
- convergence comfort
- depth direction
- how far subjects and backgrounds sit relative to the screen
They are not meant for final output. Toggle them off when saving clean previews.
The Output & Encoding section controls final output format, stereo layout, codec, quality, HDR behavior, and file compatibility.
The new codec presets make encoding easier by matching settings to common goals.
| Output Format | Use Case |
|---|---|
| Full-SBS | Best side-by-side quality, full width per eye |
| Half-SBS | Smaller file, common TV/player support |
| VR / SBS | VR-style side-by-side output |
| VR180 Equirect Top-Bottom | VR180 headset output |
| VR180 Equirect Side-by-Side | VR180 headset output |
| Red-Cyan Anaglyph | Preview or red/cyan glasses |
| Passive Interlaced | Passive 3D displays |
| Stereo Output | Meaning |
|---|---|
| SBS | Normal stereo pair |
| Left eye only | Export only left eye |
| Right eye only | Export only right eye |
| Both eyes separately | Export left and right as separate files |
Best for maximum playback compatibility.
Use when:
- sharing with users
- playing on TVs
- uploading to common platforms
- avoiding codec issues
Usually uses safe H.264 style output.
Best for quick test renders.
Use when:
- tuning settings
- checking a short clip
- comparing presets
- testing screen plane or convergence
Recommended default for NVIDIA users.
Use when:
- rendering final clips
- you want good quality and speed
- NVENC is available
Best for higher quality NVIDIA renders.
Use when:
- making showcase clips
- quality matters more than speed
- larger files are acceptable
Best when file size matters.
Use when:
- storing many renders
- uploading large clips
- reducing file size
- your player supports HEVC
Best for high-resolution Full-SBS renders.
Use when:
- rendering 4K
- rendering Full-SBS
- making high-quality headset or archive output
Best for depth-only or utility outputs.
Use when:
- exporting grayscale depth videos
- creating intermediate files
- testing depth pipeline output
Best for high-quality masters.
Use when:
- preserving quality
- creating edit-friendly masters
- storage size is not a concern
Best fallback when hardware encoding fails.
Use when:
- NVENC fails
- AMF fails
- QSV fails
- no hardware encoder is available
Use Custom when you want manual codec control.
Advanced users can tune:
- codec
- CQ / CRF
- bitrate
- preset
- container
- HDR
- audio
- pixel format
| Goal | Recommended Preset |
|---|---|
| First test | Fast Preview - Quick Test |
| Most final NVIDIA renders | Balanced Final - NVIDIA |
| Best NVIDIA quality | High Quality Final - NVIDIA |
| Small file size | Small File - HEVC |
| 4K Full-SBS | 4K / Full-SBS High Quality - HEVC |
| Maximum compatibility | Compatibility Mode - Plays Everywhere |
| Depth utility output | Depth Map Output - Fast Safe |
| Archive quality | Archive Master - Large File |
| Hardware encoder fails | CPU Compatibility - Slow |
| Manual tuning | Custom |
- H.264 is safest for compatibility.
- H.265 / HEVC gives smaller files but may not play on older devices.
- AV1 is efficient but requires newer hardware and player support.
- NVENC is recommended for NVIDIA GPUs.
- AMF is used for supported AMD GPUs.
- QSV is used for supported Intel GPUs.
- CPU encoding is slower but reliable.
Enable HDR10 preservation only when:
- your source is HDR10
- your output codec supports it
- your playback device supports HDR
Disable it when:
- source is SDR
- output looks washed out
- playback device has poor HDR support
VR180 output is for stereoscopic headset viewing.
Supported modes can include:
- VR180 Equirect Top-Bottom
- VR180 Equirect Side-by-Side
| Preset | Use Case |
|---|---|
| 2048x1024 per eye | Fast tests |
| 3072x1536 per eye | Balanced VR |
| 3840x1920 per eye | High-quality VR |
| 4096x2048 per eye | Very high quality |
| 5760x2880 per eye | Heavy showcase renders |
| Preset | Use Case |
|---|---|
| 1280x720 | Fast working render |
| 1920x1080 | Balanced quality |
| 2560x1440 | Cleaner source before VR warp |
- Use short test clips.
- Keep pop-out moderate.
- Avoid strong edge violations.
- Use Edge Repair High for final clips.
- Test in a headset before full render.
Clip Range lets you render only part of a video.
Use it for:
- testing settings
- checking difficult scenes
- comparing presets
- avoiding long failed renders
Example:
Start: 00:01:20
End: 00:01:35
The FPS / Upscale tab can improve smoothness and resolution.
It can:
- extract frames
- run RIFE interpolation
- run Real-ESRGAN upscaling
- rebuild video
- optionally restore audio
| Multiplier | Example |
|---|---|
| x2 | 30 FPS to 60 FPS |
| x4 | 30 FPS to 120 FPS |
| x8 | 30 FPS to 240 FPS |
Higher multipliers are smoother but require more processing.
Use it for:
- old movies
- low-resolution clips
- DVD sources
- cleaner 3D output
Best for:
- long videos
- lower memory use
- reliability
Best for:
- strong GPUs
- faster processing
- shorter clips
- high-performance workflows
VD3D Live is a real-time 2D-to-3D system.
It can work with:
- screen capture
- video players
- games
- cameras
- capture cards
Use the current VD3D shift direction:
FG shift: -5 to -10
MG shift: -0.5 to -2
BG shift: +2 to +5
Starter:
FG/MG/BG: -6 / -0.8 / +2.2
Stronger:
FG/MG/BG: -8.5 / -1.2 / +3.5
- Lower inference resolution for speed.
- Lower Depth FPS for stability.
- Use FP16 on CUDA.
- Use smoothing to reduce flicker.
- Avoid capturing the preview window on the same monitor.
Recommended backend for VD3D.
Best for:
- depth estimation
- 3D rendering
- interpolation
- upscaling
- NVENC encoding
Useful on Windows for AMD and Intel GPUs.
DirectML may be slower than CUDA and some operations may fall back to CPU.
Available on supported Linux systems and supported AMD GPUs.
CPU fallback works but is much slower.
| GPU / Backend | Encoder Options |
|---|---|
| NVIDIA | h264_nvenc, hevc_nvenc, av1_nvenc |
| AMD | h264_amf, hevc_amf, av1_amf |
| Intel | h264_qsv, hevc_qsv, av1_qsv |
| CPU | libx264, libx265, libaom-av1, libsvtav1 |
Check:
- depth inversion
- eye order
- foreground shift direction
- old presets
- depth map convention
Try:
- enable Layered Depth Warp
- increase BG Push
- lower Depth Pop Gamma slightly
- increase Background Shift carefully
- increase Screen Depth slightly
- use Depth Blender
- try a different depth model
Do not only increase 3D Strength. A flat depth map needs better depth shaping or blending.
Try:
- increase BG Push
- enable Depth-Order Warp
- use Wide / Deep Scene style
- use Deep Background style
- check depth map background detail
- reduce Subject Plane Lock Width if it protects too much
Try:
- increase Subject Lock slightly
- increase Subject Plane Lock
- use Subject Plane Lock Width around 0.12
- keep Screen Depth closer to neutral
- lower convergence if the stereo volume slides too much
Try:
- Edge Repair High
- Edge Masking On
- Feathering On
- lower Layered Depth Warp Strength
- lower Max Pixel Shift
- reduce Foreground Shift
- reduce BG Push
Try:
- reduce Max Pixel Shift
- reduce 3D Strength
- increase Depth Comfort
- lower Parallax Balance
- reduce convergence
- use VR Comfortable style
- render a short test before final
Try:
- lower FG Pop
- increase Depth Pop Gamma
- use Foreground Curvature lightly
- adjust Midground Shift closer to Foreground Shift
- use Depth Blender
- reduce overly aggressive Background Shift
Check:
- labels are registered for translation
- combo boxes rebuild translated display text
- combo itemData stores English preset keys
- PreviewPanel has refresh_labels
- JobQueueDock stores status keys instead of translated status text
- dynamic log history does not need to retranslate after it is printed
FG Shift: -6.4
MG Shift: -2.1
BG Shift: +2.2
Max Pixel Shift: 0.053
Parallax Balance: 0.35 to 0.45
Depth Pop Gamma: 0.94 to 1.05
BG Push: 1.03
Subject Lock: 0.40 to 0.65
Subject Plane Lock: 0.20 to 0.30
Subject Plane Lock Width: 0.12
Layered Depth Warp: 0.35 to 0.50
Edge Repair: Balanced or High
FG Shift: -8.9
MG Shift: -2.5
BG Shift: +2.9
Max Pixel Shift: 0.071
Parallax Balance: 0.40
Depth Pop Gamma: 0.88
BG Push: 1.08
Subject Lock: 0.65
Subject Plane Lock: 0.25
Subject Plane Lock Width: 0.12
Convergence Strength: 0.015
Dynamic Convergence: On
Layered Depth Warp: 0.50 to 0.65
Edge Repair: High
FG Shift: -9.0
MG Shift: -3.0
BG Shift: +2.4
Max Pixel Shift: 0.071
Parallax Balance: 0.35
Depth Pop Gamma: 1.00
Subject Lock: 0.80 to 0.95
Subject Plane Lock Width: 0.12 to 0.15
Layered Depth Warp: 0.35 to 0.50
Edge Repair: High
Edge Masking: On
Feathering: On
FG Shift: -9.9
MG Shift: -3.0
BG Shift: +3.3
Max Pixel Shift: 0.071
Parallax Balance: 0.70 to 1.00
Depth Pop Gamma: 0.87
BG Push: 1.08 to 1.12
Subject Lock: 0.70 to 0.90
Subject Plane Lock Width: 0.12
Layered Depth Warp: 0.65
Edge Repair: High or Showcase
- Start with beginner controls.
- Use preview before rendering.
- Render short clips first.
- Use Shift Heatmap to inspect stereo stress.
- Use Anaglyph to check real depth feeling.
- Use Layered Depth Warp when backgrounds are flat.
- Keep Subject Plane Lock Width around 0.10 to 0.14.
- Use Edge Repair High for final showcase clips.
- Use Balanced Final or High Quality Final for NVIDIA output.
- Use Compatibility Mode when sharing widely.
- Do not overuse convergence to create depth.
- Use depth maps with strong subject and background separation.
- Blend depth maps when one model is not enough.
Balanced depth usually looks more cinematic than extreme pop-out.
For updates, documentation, and releases:
- GitHub repository: VisionDepth3D
- Official VisionDepth3D website
- Community feedback and issue reports are welcome
VD3D continues to improve depth quality, stereo stability, render speed, preview tools, language support, encoding presets, and user workflow.