BlenderBlender and AI

Making a 3D character with Claude Code, part 4: chaining Asset Shelf poses and rendering

Set Asset Shelf body poses and faces frame by frame to make your own wave and bow, chain them to NLA tracks with Push Down, render the result as a video, and move the camera.

AZTechnology17 min read

한국어English

Based on: Blender 5.2 (English UI, default keymap) and the Blender 5.2 manual (October 2026). Numbers and names are from this series' character file. All screen images are real Blender 5.2.1 screens. The red numbers and outlines were added for explanation.

In part 3 you turned on, moved, and blended the existing NLA tracks to build a scene. You didn't set a single key.

In part 4 you make new movement yourself. You set keys by applying Asset Shelf poses frame by frame, add faces, chain the result to the existing tracks with Push Down, and render. This is the last part of the series.

By the end of this part you will be able to:

  1. Apply Asset Shelf poses frame by frame to make your own motion, and chain it to NLA tracks.
  2. Key the face before the body so the expression doesn't get reset.
  3. Render the scene as a video and move the camera.

1. Chaining Asset Shelf poses

Use this when you want new movement instead of the ready-made motions. The idea is stepping stones: set a pose at certain frames, such as the neutral standing pose at frame 1 and the hand-raised pose at frame 30, and Blender fills in the movement between them.

The Asset Shelf is a shelf under the 3D Viewport that lines up pose thumbnails. Each thumbnail on the shelf is a one-frame pose, like a single photo. Click a thumbnail and that pose goes onto the character.

Pose previews in two rows: 7 body poses on top and 4 facial expressions on the bottom Preview thumbnails of the 11 poses on the Asset Shelf. Top row body poses, from the left: neutral standing, hand raised, deep bow, bow end, ready to draw the sword, slash end, walk step. Bottom row faces: neutral, eyes closed, smile, surprise (Blender render)

CatalogPose names
saya/poses (body)pose_attention (neutral standing), pose_wave_hi (hand raised), pose_bow_deep (deep bow), pose_bow_end (bow end), pose_draw_ready (ready to draw the sword), pose_slash_end (slash end), pose_walk_step (walk step)
saya/face_poses (face)pose_face_neutral (neutral), pose_face_blink (eyes closed), pose_face_smile (smile), pose_face_surprise (surprise)

2. Setup

Starting point: Continue from the copy you saved in part 3. If you don't have one, save 05_scene.blend under a different name as in "Before you start" in section 1 of part 3, and use that. First turn off every LIB_* track you turned on. If they stay on, they mix with the poses you set and get confusing. The pose names in this part are from the series' file. Your 05_scene.blend may have different track and pose names, so after opening the shelf in step 4, check the pose names first.

Try it:

  1. Enter 120 in End on the right of the Timeline. At 30 fps, that is 4 seconds.
  2. Click the character's armature in the 3D Viewport to select it, then press Ctrl+Tab to enter Pose Mode.
  3. In the 3D Viewport header, turn on View ‣ Asset Shelf. A shelf of thumbnails appears along the bottom of the 3D Viewport.
  4. At the top left of the shelf there is only one tab, All. The pose catalog names don't show up as tabs right away. Type pose_ in the search field at the top right of the shelf and only the 11 poses remain.

The Asset Shelf opened under the 3D Viewport in Pose Mode: the Pose Mode indicator, the All tab, the search field with pose_, and a shelf with 4 face and 7 body thumbnails, with outlines 1 to 4 ① Pose Mode ② All tab ③ Search field (pose_) ④ Pose thumbnails. The four on the left are faces; the rest are body poses. The shelf in the image was enlarged so the names show (screenshot, Blender)

In the middle of the Timeline header, the round button to the left of the play buttons is Auto Keying (Timeline image ② in section 4 of part 3). When it is on, a key is set at the current frame every time you move a bone, like a record button. When you click a pose on the shelf, though, keys did not go in as expected (box in section 3). This guide keeps Auto Keying off and sets keys by hand.

3. Setting poses

The basic order for setting one pose is deselect → click the thumbnail → select all → insert key. If you click a thumbnail with no bones selected, the pose goes onto the whole body; if even one bone is selected, it goes only onto the selected bones.

Alt+A → click thumbnail → A → I

Try it:

  1. Click frame 1 in the Timeline.
  2. With the mouse over the 3D Viewport, press Alt+A to deselect all bones.
  3. Click pose_attention on the shelf. One click changes the pose.

Two camera views of the character. Left: the starting neutral standing pose. Right: the right hand raised after clicking pose_wave_hi on the shelf Left: before the click. Right: after clicking pose_wave_hi. At this point only the pose has changed; there is no key yet (screenshot, Blender)

  1. With the mouse over the 3D Viewport, press A (select all bones) → I (insert key). A diamond (◆) key appears in the Timeline. Check for the ◆ before moving on to the next pose.
  2. Go to frame 30 and repeat steps 2 to 4 with pose_wave_hi.
  3. Go to frame 60 and repeat steps 2 to 4 with pose_attention.
  4. Go back to frame 1 and press Space. She moves from neutral → hand raised → neutral.

A → I keys the face too: When A selects every bone, it also selects the head control bone that holds the face values (ctl_head). I keys location, rotation, and scale, and also the bone's custom properties. That happens when Custom Properties is turned on under Default Key Channels in the Keyframes panel of Edit ‣ Preferences ‣ Animation. So every time you key a body pose, the face values at that frame (smile, blink, surprise) are keyed too. In scenes that use faces, key the faces first and the body later (section 6). That way the body keys write back the face values already at that frame, so the faces stay almost unchanged.

When Auto Keying is on: In this file, with Auto Keying on, running the same command as a shelf click produced no ◆ in the starting state with no keys. Once keys already existed, only some of the changed values got keys (27 of 260 keyable curves). To keep a whole-body pose intact, don't rely on Auto Keying; insert keys with A → I as in step 4.

Playing a file with the same poses fully keyed, with every track off, through frames 1→30→60, the right hand rises from about 0.36 m to 0.67 m and comes back down.

Face poses change only the face values (closing the eyes, smiling, looking surprised) and leave the body alone. Key faces with I over the value, not A → I. With A → I, the body pose at that frame gets keyed too.

Starting point: If your file already has the body keys from section 3, start from a new copy. If the body keys come first, the face values stored in them override the new faces (box in section 3).

Try it:

  1. Type pose_face in the shelf search field to keep only the 4 face thumbnails.
  2. At each frame in the table below, click the face thumbnail and key it as described in "Keying a face" below.
FrameFace to click
1pose_face_neutral
25pose_face_smile
40pose_face_neutral
43pose_face_blink
46pose_face_neutral

From the neutral face at frame 1, she smiles gradually up to frame 25, and the smile fades by frame 40. To keep her smiling, use the smile instead of neutral at frames 40 and 46.

Keying a face: After clicking a face thumbnail, click the head control bone that holds the face values (in this file ctl_head, the yellow circle in front of the face) to select it. With the mouse over the 3D Viewport, press N to open the sidebar on the right. Choose the Item tab, then click the collapsed Properties panel near the bottom to expand it. Hover over each of the three values, blink, smile, and surprise, and press I each time. When a value field turns yellow, the key is in. When you've keyed the faces, press Alt+A to deselect.

The 3D Viewport sidebar with outlines 1 and 2 on the vertical Item tab on the right and the collapsed Properties panel under Transform ① Item tab ② Properties panel. It starts collapsed; click to expand it and see blink, smile, and surprise. The ARP and Forest Studio tabs on the right of the image are separately installed add-ons and may not be on your screen (screenshot, Blender)

5. Blending halfway and flipping left and right

  • Right after you click a thumbnail, an Apply Pose Asset panel (the panel for adjusting the last operation) appears at the bottom left of the 3D Viewport. Expand it and set Blend Factor to 0.5 to blend the pose halfway with the current pose; turn on Apply Flipped to swap left and right (the left hand instead of the right). The blend is with the current pose at that frame, so blending in the same pose as the current one changes nothing.
  • If you don't see the Apply Pose Asset panel, press F9 right after clicking the thumbnail. The same panel opens. If that is still awkward, drag the thumbnail to blend, as in the next line.
  • Drag a thumbnail from left to right to set the amount of blending while you watch. Release the mouse button to confirm. You can also right-click a thumbnail and choose Blend Pose.

6. Exercise: your own wave and bow

Starting point: Start from a new copy that has finished the setup in section 2. Every track is off, you are in Pose Mode, and no bones are selected. End is 120.

Try it: Key all the faces first, then the body poses. Key faces with "Keying a face" in section 4 (I over the value), and the body with the order in section 3 (Alt+A → click thumbnail → A → I). If you do it the other way round, the face values stored in the body keys (default smile 0.4) make the frame 25 smile drop off right away at frame 29 (box in section 3).

Step 1, faces (saya/face_poses):

FrameClickMeaning
1pose_face_neutralStarting face
25pose_face_smileBiggest smile
40·43·46neutral → eyes closed → neutralThe smile fades, then a blink

Step 2, body poses (saya/poses):

FrameClickMeaning
1pose_attentionStart
15pose_wave_hiHand up
22pose_attention → Blend Factor 0.4 (panel in section 5)Lower only 40% toward neutral (the wave)
29pose_wave_hiUp again
50pose_attentionHand down
75pose_bow_deepDeep bow
85pose_bow_deepBrief hold at the lowest point
100pose_bow_endStands up
115pose_attentionEnd

Frame 22 starts from the hand-raised pose (the frame 15 key). Blending in only 40% of neutral lowers the hand a little. This number sets how wide the wave is. After entering 0.4, check that the right hand has come down, then key it with A → I.

When you're done, play from frame 1 with Space. Keyed in this order and measured, the smile value (smile) rises from 0 at frame 1 to 1.0 at frame 25, then fades to 0.83 at frame 29 and 0 at frame 40. Her eyes close once at frame 43, and the neutral face holds from frame 46 to the end.

The Timeline header and key row, with outline 1 on a row of diamond keys at frames 1, 15, 22, 25, 29, 40, 43, 46, 50, 75, 85, 100, and 115 ① The ◆ keys after all face and body keys are in. The Timeline is shown with every bone selected (A) (screenshot, Blender)

Five camera views of the character: hand raised at frame 15, hand slightly lowered at frame 22, rising again at frame 25, eyes closed at frame 43, and bowing at frame 85 The result of the exercise. From left: frames 15, 22, 25, 43, and 85. The image hides the bones and camera lines (screenshot, Blender)

To adjust the pose at a frame a little, click a circle or box control bone and use R (rotate) or G (move). Afterward, with that bone still selected, press I to key it again. If Auto Keying is on, the key updates automatically. To change a face later, remember that the body key frames in that range hold face values too, so key the face again with I over the value at those frames as well. To delete a key, click the ◆ in the Timeline and press X.

7. Chaining to tracks with Push Down

The keys you set live in the orange row at the top of the NLA (the active action). This row overrides every track. Leave it as is, and the bones you keyed won't move even when you turn tracks on.

Try it:

  1. In the Nonlinear Animation editor, click the Push Down button (named Push Down Action in the manual) at the right of the orange row at the top (pose_user). The small dots in the row are the keys you set.

The orange row pose_user at the top of the NLA Editor, with key dots along it and outline 1 on the Push Down button at the right end ① Push Down button (screenshot, Blender)

  1. Your keys move down into a strip on a new track (pose_user) just below the orange row, and the orange row empties to <No Action>.
  2. Click the new strip and, under N → Strip tab, change Extrapolation to Nothing. A strip pushed down with Push Down is set to Hold (keep the end pose both before and after). Left like that, it keeps overriding the motions on the tracks below even after it ends. The pushed-down strip also holds face keys, so while it plays, it overrides the face tracks below too.

Two views of the NLA Editor, right after Push Down and after the fix. Top: the new pose_user track and Extrapolation Hold. Bottom: the sidebar changed to Extrapolation Nothing. Outlines 1 to 3 ① The track created by Push Down ② Starting value Hold ③ After changing it to Nothing. In the top view, the orange row has emptied to <No Action> (screenshot, Blender)

  1. Move this strip as in section 5 of part 3, and chain it with a ready-made motion such as LIB_bow using Blend In. Have the next motion start around the frame where the new strip ends.
  2. Save with Ctrl+S.

In this file, with the strip ending at frame 115 and LIB_bow turned on from frame 116, the head stayed put even at frame 150 while the strip was set to Hold (the bow was overridden). After switching to Nothing, she bowed.

If you only want to store the keys, use Stash in the Action Editor header of the Dope Sheet. They go into a muted track.

8. Rendering

Starting point: Continue from the scene you made in section 4 of part 3 or in section 7 of this part.

Try it:

  1. In the Output panel of the Output Properties tab, change Media Type to Video. The default container and codec (Matroska, H.264) are fine. Detailed settings are in Blender beginner chapter 11. End in the Frame Range panel of the same tab is the same value as End in the Timeline.

The Output Properties tab in the Properties editor, with outlines 1 to 3 on the Output tab icon, End in Frame Range, and Media Type Video in the Output panel. The output path field is blurred ① Output Properties tab ② End (132 in the image; set it to fit your scene) ③ Media Type: Video. The output path field is hidden (screenshot, Blender)

  1. Choose Render ‣ Render Animation (Ctrl+F12).
  2. First render at the file's existing 960x540 to check it. If you like it, raise the Resolution and render again.

Moving the camera: The camera in this file is tied to follow a look-at target. Turn that off first, and then it moves freely.

  1. Select the camera CAM_user, and in the Object Constraint Properties tab, turn off the eye icon of look_at_target (Track To).

The Object Constraint Properties tab in the Properties editor, with outlines 1 and 2 on CAM_user's look_at_target constraint and the eye icon next to its name ① Object Constraint Properties tab ② The eye icon of look_at_target. Click it off and the camera stops following the look-at target (screenshot, Blender)

  1. With the mouse over the 3D Viewport, press numpad 0 to enter the camera view.
  2. Under N → View tab, turn on Lock Camera to View, then orbit the view to frame the shot.
  3. When it's framed, turn Lock Camera to View off.

9. Common mistakes and fixes

SymptomCauseFix
I turned on a track and nothing movesThe checkbox is off / the orange row has keys on that boneTick the checkbox / Push Down in section 7
I clicked a pose, but nothing moves on playbackNo key went inAfter clicking, insert keys with A → I
The motion chained after the Push Down strip doesn't showThe new strip keeps holding its end pose with HoldSet the new strip's Extrapolation to Nothing
The smile drops off right at the next body keyThe body key (A → I) also stored the face values at that frameKey the faces first and the body later (section 6)
Changing Blend Factor leaves the pose the sameThe same pose as the current one was blendedPick a pose different from the current one (for example, neutral) and blend
The Asset Shelf is emptyNot in Pose Mode / other text is left in the search fieldSteps 2 to 4 of section 2
The shelf has no pose catalog tabsAt first there is only the All tabType pose_ or pose_face in the search field
With Auto Keying on, clicking a pose gives no keys or only someA click doesn't key the whole poseInsert keys with A → I
A pose I fixed by hand snaps back when I change framesThat bone has keys on a track that is turned on or in the orange rowTurn the track off, or key the fixed pose with I

10. Wrap-up

To sum up: the drawing is one Higgsfield image, the model, rig, and motions are Python scripts written by Claude Code, and only the walk is from Mixamo. What the author did was look at the result of each stage, send the next request, and make the final call.

There are limits. The same kit is also being used for a character about 5 to 6 heads tall, but sleeves and skirt still poke through each other, and it is still in the 6-point range. The longer the body and the more the clothes overlap, the harder it is to get right with numbers alone.

In one line: one drawing + a data file + scripts for an easy-to-edit character, then NLA tracks and Asset Shelf poses to build scenes.

To go through the series again from the start, return to Part 1: from one drawing to an easy-to-edit model.

References

The full guide, with the reusable kit and pose library, is a book on Gumroad: https://azblender.gumroad.com/l/lowpoly_character

KEEP EXPLORING

Read next

A frame from a low-poly video made with an AI agent and Blender: drones rising over a grass field (Blender render)Blender
Blender and AI17 min read

AI and Blender: Start Small and Work Together

When you hand Blender work to an AI, split the job, read before running, check with a picture and numbers, and save with version numbers. This guide walks through that order with a real production example.

The Scripting workspace: a Python script that builds a staircase from five cubes in the Text Editor in the middle, the result in the 3D Viewport at the top left, and the log of executed commands in the Info editor at the bottom left (Blender screenshot)Blender
Blender and AI11 min read

Connecting Blender and AI: 3D work through MCP and scripts

A plain comparison of the two ways AI can drive Blender: Python scripts and an MCP connection. It also covers how they work, a first exercise, and safe working habits.

Low-poly scene of interceptor missiles hitting a drone swarm over the desert, with the title Air Defense (Blender render)Blender
Blender and AI9 min read

Low-poly war animation with Claude Code, part 1: one day, and who did what

With almost no Blender experience, I made five 40-second 3D animations in one day by giving Claude Code instructions in plain words. This part covers that day and splits out what the AI did and what I did.

Low-poly scene with launchers and vehicles lined up in front of orange desert mountains, and a drone formation, intercept explosions and white smoke trails in the sky (Blender render)Blender
Blender and AI12 min read

Low-poly war animation with Claude Code, part 2: the five films, results first

The five low-poly war videos I made in one day by giving Claude Code instructions in plain words, shown first: one key frame, three lines of description and one key number for each.

Farmland with green hills and mountains. A tank in the front right aims its gun, and explosions go off near a distant enemy tank column. The title Tanks & Artillery at the bottom left (Blender render)Blender
Blender and AI14 min read

Low-poly war animation with Claude Code, part 3: install Claude Code and let the AI set up the rest

Install Claude Code on Windows and log in, hand the checking and installing of Blender, ffmpeg and the Python libraries to the AI, and confirm you're ready with one picture of a cube.

Theater tab of a dark web app, with four cards in a grid: a desert air-defense scene, a submarine launch, a tank scene and a fighter jet scene (tracker screenshot, labels in Korean)Blender
Blender and AI13 min read

Low-poly war animation with Claude Code, part 4: the first prompt, with your plan on one page

Take the real thirteen-line first prompt apart into five groups, learn to give specs and quantities in numbers, get a plan first in plan mode, and fill in a first-prompt template of your own.