Where Anima falls short

Last reviewed: 2026-10-12 · Last updated: 2026-10-13 · Change: minor (path-editing layer bounds, #1220; issue, PR and catalog references gap 21: percentage guides and effect colour labels in; earlier change: re-audited against upstream/main 470ea7a3; ranked gap list from the 10 October re-measure otherwise unchanged) · Target: Adobe After Effects 2026 (26.5.0.89)

An honest assessment of how far Anima is from being a real replacement for After Effects, and the work that closes the gap. This document is meant for contributors and agents choosing what to work on. target-app-parity.md holds the numbers and how they were measured; the ROADMAP summarises both. Estimates marked "≈" are judgements from the evidence listed, not measurements. Hours are Opus 5.5 agent wall-clock hours, calibrated in target-app-parity.md. Update this file when the evidence changes.

Two different questions

Feature breadth ≈ 91% answers does each After Effects feature exist? Every one of After Effects' effects exists by name, along with the menus, panels, Classic and Advanced 3D, tracking, expressions, scripting and export. The self-graded 92-item checklist (appendix of target-app-parity.md) says ≈ 99%; measuring against After Effects 26.5 itself, rather than our catalogue, finds what that checklist never listed: the 26.2–26.5 features (Object Matte, Substance materials, ACES 2.0, Proportional Scrubbing, paste SVG/AI as shapes…), the 621 bundled animation presets, project and template formats, and hardware encoding.

Ready for real work ≈ 45% answers can someone who uses After Effects for a living do real work in Anima? It is still estimated, not measured. Four things hold it back: projects we can't open (.aep), behaviour nobody has checked against After Effects (one live comparison so far: scalar ease, 168 samples), reliability outside macOS, and the third-party plug-ins and presets professional projects depend on. It sits higher than the 5 October estimate (30–50%) because 179 issues were closed with regression tests between 5 and 10 October.

By dimension

Every known gap, ranked

Alpha blockers: none. Every core workflow passes the alpha gate; the entries below are what stands between alpha and beta (gap 1 decides beta). The rough edges the gate lists live in gaps 5, 8, 9, 10, 11 and 17.

Ranked by user impact × how many users hit it. Each entry: what's missing, evidence, impact, estimate, the workstream (G1–G9, below) and the parity doc it belongs to. Small single-issue bugs are grouped by area; the issue tracker has the rest.

Where we're going: workstreams in priority order

Each workstream lists what "done" means, so progress can be measured rather than self-graded.

G1. Measure fidelity against After Effects

The most important missing piece: it turns every other estimate here into a measurement.

• Build a corpus of original test projects (our own content, no Adobe assets), one or more per feature id in plan/aftereffects/feature-catalog.md: keyframe interpolation of every kind, blend modes, track mattes, each effect at default and at non-default settings, text animators, expressions, 3D, motion blur, time remapping.
• Drive After Effects through ExtendScript (see CLAUDE.md) to record property values at sampled times and render reference frames; render the same projects in Anima headless.
• Compare values exactly and frames with a perceptual metric; publish a per-feature fidelity score and a fidelity column in target-app-parity.md.
• Clean room: After Effects output stays local in plan/aftereffects/ref/ (gitignored). Commit only our projects, the harness and the scores, never After Effects frames.
• Done when: every P0 and P1 feature has at least one corpus project, scores are reproducible from one command, and target-app-parity.md reports breadth and measured fidelity side by side.

G2. Real-user reliability on every platform

• Fix user issues as they come in, each with a regression test, and answer the reporter (#41–#47 and #63–#68 are handled, as are the Windows report that the font menus missed installed fonts and the reports of 6 October, #85, #88, #89, #93, #103, #106 and #117, and of 7 October, #198, #201, #203–#206, #208, #209 and #212, and of 8 October, #225, #226 and #229–#231; #227 is handled except multi-monitor support; Wayland file drops wait on winit).
• Run the headless snapshot and control-channel checks on Linux (X11 and Wayland) and Windows, not only macOS. Interactions that only fail with real input (docking drags, viewer pan, drag-and-drop import, inline rename) need scripted input tests.
• Path editing (#1220): selected shape and mask paths suppress the layer transform box while their points are edited. ui_viewer regressions cover Pen-family tools, Selection vertex dragging and undo, restoration of layer controls, and layers without visible paths.
• Regression evidence for #66: scripted pointer tests in ui_viewer and ui_text_edit, checked headlessly on Linux, cover viewer clicks, marquee selection and text picking when layers are hidden or excluded by solo. They also check hidden/locked solo layers, and that solo still excludes the other layers while the soloed layer is outside its active time range (as in After Effects, so nothing is picked there).
• Regression evidence for #67: scripted input tests in ui_project_delete, checked headlessly on Linux, cover Delete/Backspace on Project items, multiple selection and undo. They check that an unrelated Timeline layer stays, an empty Project selection does nothing, viewer focus still deletes layers, and typing or dialogs do not delete items.
• Regression evidence for #484: the shift_delete_* input tests in ui_project_delete, checked headlessly on Windows, press Shift+Delete both as a key and as the Cut event Windows sends for it. Items in use go without the prompt, with their layers, in one undo step, for single items, multiple selections and folder contents. Nothing happens with an empty selection, in other panels, while typing or with a dialog open, and the next Delete still asks.
• Regression evidence for #612: the ui_gradient_value and ui_gradient_editor input tests, checked headlessly on Windows, open the Gradient Editor from the Timeline's "Edit Gradient…" on a shape layer's Gradient Fill. They check that OK sets the edited stops (added, dragged, recoloured in the picker, down to a single colour stop) in one undo step, at the composition time the editor opened (also on offset and stretched layers), and that OK without a change, Cancel and Escape change nothing; Escape with the picker open closes only the picker. On an animated gradient only the keyframe at that time changes, with a warning when its stop counts no longer match a neighbour's. Delete never removes the last stop, a layer or a stop while typing; a stop moved past another by its Location keeps its selection and fields; a layer deleted while editing is reported; and prop.editGradient is an error for a target that isn't a gradient. Unit tests in keyframe and gradient_editor cover per-track sampling (no double opacity), the typed one-stop round trip, and stops staying in order, and selected, while they move. The ui_gradient_handles tests, checked headlessly on Windows, drag a selected Gradient Fill's start and end points in the Composition viewer: one undo step, the layer stays, a point inside a moved and scaled group converts to the group's space, and the handles show only for a selected gradient.
• Regression evidence for #1011: the ui_text_range_handles input tests, checked headlessly on macOS, drag a text Range Selector's End bar in the Composition viewer: one undo step, Start and the layer stay, and on an animated End a keyframe at the current time. The bars show only for a selected selector or a selected animator (or a property in it), with the Selection tool and layer controls on. Unit tests in text place the bars on unit boundaries (words and Characters Excluding Spaces skip spaces, inside a ligature, on the second line, right to left) and invert them; render covers the value a drag writes (less the Offset, whole units in Index). Not done: dragging Offset, bars on vertical text and on text on a path.
• Regression evidence for the toolbar's gradient swatches (#612, reported on Discord): the ui_toolbar_gradient input tests, checked headlessly on Windows, click a Linear Fill swatch and a Radial Stroke swatch on a selected shape layer and get the Gradient Editor on its Gradient Fill / Gradient Stroke Colors (OK saves); without such a gradient the swatch still opens the options, and the words, Solid swatches and Alt-click are unchanged.
• Regression evidence for the reports of 6 October, checked headlessly on Windows: real pointer drags in ui_drag_drop (Project items into the Timeline between layers and at a time, and onto the viewer; an effect onto a layer in the viewer: #85, #88, #89), ui_render_queue (a template chosen from a menu moved to fit the window, #117), angle_revolutions_scrub_and_take_typing (#93), audio_lock_and_shy_switches_keep_the_frames and preview_with_audio_shows_every_frame_and_sounds_once_cached (#103), and a_frame_whose_render_panics_is_released (#106).
• Follow-up angle-input regressions in ui_angles (#141), checked headlessly on Linux and Windows, verify that Escape cancels edits and non-finite entries leave the value and undo history unchanged in Effect Controls, Timeline and Properties. Valid entries still commit on Enter or click-away.
• Regression evidence for the reports of 7 October, checked headlessly on Linux: ui_viewer full_resolution_frames_wider_than_the_texture_limit_fit (#201), the desktop app's no_arguments_start_an_empty_project (#204) and device_limits_fit_the_adapter (#198), audio_spectrum_follows_time_through_the_layer_cache (#209), preview_with_audio_sounds_once_rendering_keeps_up (#208), ui_timeline_shapes (the Add menu and Change Renderer, #206; deselecting below the layers, Alt+Shift+P and the toolbar's stroke, #205; linked Mask Feather, #203), dragging_in_the_empty_area_box_selects_items and selecting_a_mask_selects_its_points (#203), and input_keys_ignore_later_effects (#212).
• Regression evidence for the reports of 8 October, checked headlessly on Windows: a_short_project_panel_draws (#231), viewer_press_prefers_a_selected_layer_under_the_pointer (#230), match_system_uses_the_system_language_where_there_is_a_catalog (#229) and apps/anima-web/tests/page.mjs (#226, #225); the #225, #226 and #231 reproductions also pass in the dev web build in headless Chrome 154 (device pixel ratio 2, a rejected file read, GPUDevice.destroy()).
• Regression evidence for the Turbulent Displace, Wave Warp and Bulge pinning of #227: edge_pinning_displaces_all_of_a_centred_shape_layer (render) and edge_pinning_on_a_shape_layer (GPU against CPU).
• Regression evidence for the effect points of #227 (shape and text layers measure effect points from the top-left of their comp-sized bounds, so Twirl, Bulge, CC Lens, Ripple and Turbulent Displace's Offset default to the layer centre): default_effect_points_sit_at_the_centre_of_a_shape_layer and layer_parameters_see_a_whole_shape_layer (render), effect_points_on_a_shape_layer (GPU against CPU), and the schema 1 → 2 conversion in opening_a_schema_1_project_moves_shape_layer_effect_points_into_effect_space (engine) and effect_points_move_into_effect_space_unless_left_at_their_default / puppet_paint_liquify_and_roto_data_move_into_effect_space (effects); see architecture.md §3 for what the conversion keeps.
• Regression evidence for the interface items of #227, checked headlessly on Windows: spacebar_taps_preview_and_held_spacebar_pans_the_viewer and spacebar_drag_scrolls_the_time_graph (Spacebar previews on its release, held it is the Hand tool), right_click_in_effect_controls_shows_the_effect_menu, project_items_dropped_on_new_comp_make_a_composition and dragging_over_layer_switches_sets_them_all.
• Regression evidence for the shape layer items of #227, checked headlessly on Windows: shape_tools_draw_into_the_selected_shape_layer and ui_viewer shape_tool_draws_into_the_selected_shape_layer (a shape tool adds a group to the selected shape layer), shape_contents_copy_cut_and_paste_between_shape_layers and ctrl_c_and_ctrl_v_copy_a_shape_group_into_another_shape_layer (Copy, Cut and Paste of shape items), duplicate_with_shape_items_selected_duplicates_them_in_place and ctrl_d_duplicates_the_selected_shape_group_in_its_layer (Duplicate), and shape_tool_options_paint_new_shapes, masks_of_every_shape_tool_kind, tool_creates_mask_draws_a_mask_on_the_selected_shape_layer and fill_options_paint_the_next_shape (Tool Creates Shape / Mask, Fill and Stroke Options).
• Done when: no open bug blocks a basic workflow (import, arrange, animate, preview, render) on any of the three desktop platforms, and every bug users report gets triaged within a day.

Real-user report log (5–10 October)

Kept from the 5 October assessment's "Real-user experience" row, as evidence of what users hit and what was fixed:

Issues #41–#47 (all from the Linux AppImage 0.1.1, 5 October): panning the viewer snaps back, panels can't be resized or rearranged, drag-and-drop and double-click import don't work, layer rename gets stuck, the Layer Settings arrow does nothing, the Project panel clips, the Wayland window icon is generic. Issues #63–#68 (macOS, 5 October): scaling a layer by its handles goes wrong and can stick at 0, two 3D compasses, the viewer lags while a layer is dragged (a frame renders in 20 ms but showed after ≈ 100 ms, behind RAM preview prefetch), the Discord and other links do nothing, Delete doesn't delete in the Project panel, hidden layers can be selected in the viewer. All but the last two were fixed or confirmed fixed with a test on 5–6 October (those two have community PRs); file drops still can't work on Wayland (winit has no support). The panning, rename and arrow bugs had been fixed in v0.2.0 already; nobody had told the reporter. A Windows user (6 October, on Discord) found the font menus listed only the three bundled fonts: installed fonts were read only when a project asked for one, on every platform. Fixed with a regression test on 6 October; the menus now list every installed family and its own styles, and agents get text.fonts / list_fonts. Seven reports of 6 October (Linux .deb and Windows): Project items and files couldn't be dropped on the viewer (#85), nor effects on a layer there (#88); drops in the Timeline ignored where they landed (#89); an angle's revolutions couldn't be edited (#93); toggling Audio, Lock or Shy threw away the RAM preview and playback with audio skipped frames (#103); running out of video memory panicked every frame thread and left frames stuck (#106); choosing a Render Settings template in the Render Queue did nothing (#117: any popup menu moved up to fit the window closed on the press). All seven were fixed with regression tests on 7 October. Reports of 7 October (Linux, macOS, Windows): the app opened the demo project on every launch, so a new comp landed in it (#204); a comp wider than the GPU's texture limit (11000 × 2200) at Full resolution stopped the window drawing (#201); an Intel HD Graphics 5500 couldn't start the app (#198: most likely the window's device asked for limits that GPU doesn't have, and the failure was silent); Audio Spectrum / Waveform stood still in the viewer (#209) and preview sound waited for the whole work area to cache (#208); a layer couldn't be deselected from the time graph, a shape layer's Fill and Stroke showed only with a shape tool and a drawn shape had no stroke to change, Alt+Shift+P didn't reveal Position (#205); a mask selected in the Timeline wasn't selected in the viewer, Mask Feather had no link and took negative values, the Project panel had no selection box (#203); shape layers had no Add menu for Trim Paths and the other operations, and Classic 3D comps hid Geometry Options (#206); dragging Puppet pins re-rendered the layer under the Puppet at every move (#212). All were fixed with regression tests the same day (#219–#222); the #198 fix couldn't be checked on that GPU. Reports of 8 October (browser, Windows, macOS): an empty project at 1280 × 800 with device pixel ratio 2 crashed on its first frame, because the Project panel's scroll bar clamped its thumb to a track shorter than the thumb (#231, any short panel on every platform); Media Browser ▸ Add Files… dropped a file it couldn't read without a word (#226); after the page's WebGPU device was lost, the browser editor froze on its last frame while edits went on (#225); a press in the viewer always took the topmost layer, so a selected layer behind others couldn't be dragged (#230); the language didn't follow the system's (#229); and a list of missing After Effects behaviour (#227): drawing into, copying, pasting and duplicating shape contents, Tool Creates Shape / Mask and Fill / Stroke Options, Turbulent Displace, Wave Warp and Bulge pinning all but a shape layer's bottom-right quadrant, a Spacebar hold that previewed instead of being the Hand tool, the Effect menu in Effect Controls, dropping items on New Composition and dragging over layer switches. All were fixed with regression tests the same day (#235, #237–#241); multi-monitor support and the Project panel's empty-area menu (community PR #210) remain from #227, and effect points on shape and text layers still sit half a comp off. A list of tweaks from an Intel iMac (#290): the layer bar didn't show where extended footage runs past its source, and Enable Time Remapping then ended at the extended out point; footage items had no Interpret Footage in their menu and the work area bar no menu; Label menus named the labels without their colours; the arrow keys didn't nudge layers; a straight motion path had no Bezier handles; Delete on Time Remap deleted the layer; a mask being drawn with the Pen blacked out its layer until it was closed; a double-click on a mask point didn't select the whole mask. All were fixed with regression tests the same day; the rest of #290 had been fixed already (#221, #238, #255) or came with community PRs #210 (the Timeline's empty-area menu) and #213 (tool flyouts on a long press). A report of 9 October (#397): CC Particle World ignored the comp camera and looked flat. It now sees its particles through an active camera layer, as After Effects does, and through its own Extras ▸ Effect Camera (Distance, Rotation X / Y / Z, FOV, added) otherwise; the layer cache also missed camera and light moves for the effects that read them (Card Dance, Shatter, Card Wipe, Caustics). Fixed with regression tests the same day. Most checking happens on macOS. Localisation covers the menus in Japanese, Simplified and Traditional Chinese, and Ukrainian, and since 9 October the cataloged panels, dialogs and tooltips too (Simplified and Traditional Chinese, and Ukrainian; the Japanese rows of the panel catalog are still open); no accessibility work

G3. Stability and a green main

• Keep cargo xtask ci green on the current stable toolchain. A toolchain release that breaks the gate is a P0.
• Fuzz the inputs we don't control: project files, imported media and documents, expressions, scripts, control-channel and MCP requests. Every crash found becomes a regression test (AGENTS.md "Never crash"). Project files are fuzzed (#779): a seeded, dependency-free mutation fuzzer runs in cargo test -p anima-engine (fuzz_file_open_never_crashes, .ecproj JSON and .ecprojx XML through File ▸ Open, the first queries and a thumbnail render; longer runs via ANIMA_FUZZ_ITERS / ANIMA_FUZZ_SECS). It found one crash, deeply nested XML overflowing the stack, now refused. Imported media and documents, expressions, scripts and control-channel / MCP requests are still to do.
• Run the gate on every pull request in CI (gap 4): today no workflow runs the tests on PRs.
• Fix the open crash reports first (gap 3). Further fuzzing and crash fixes are open in PR #855.
• Done when: the gate is green on stable, runs on every pull request, and the fuzzers run regularly without new crashes.

G4. Open After Effects projects

• .aep / .aepx import is the largest single barrier to switching. It needs an owner decision on clean-room scope (whether and how the format may be studied), like the open .prproj question in plan/STATUS.md.
• A candidate for that decision (#248): py-aep, an MIT licensed Python library that reads and writes .aep files, reported by a user as working well. It can't be linked into the pure-Rust app. It could run as an external converter (.aep to .ecproj, the way ffmpeg is used only from outside), or its format knowledge could be ported. The second means reading another implementation of the format, which AGENTS.md rules out unless the clean-room scope allows it.
• Also in this area: relative footage paths and relinking when a project moves (gap 18).
• Done when: the decision is recorded and, if approved, a corpus of real-world-shaped projects opens with its comps, layers, keyframes, effects and expressions intact.

G5. Performance at real-world scale

• Benchmark projects at 1080p and 4K with many layers, heavy effects, long footage and nested comps; record preview frame times, RAM preview fill and render times on reference machines. Started 10 October 2026: cargo xtask bench renders generated 1080p / 4K scenarios (many layers with common effects, nested precomps, footage) at Full and Half and records peak memory, CPU and wall time per run, with --compare against an earlier run (docs/testing.md, "Render benchmark").
• Compare with After Effects on the same machine where possible.
• Done when: benchmark numbers are tracked over time and regressions fail a check.

G6. Media depth

• Encoder efficiency: B-frames and multi-reference for HEVC / AV1, SAO / CDEF / restoration, Opus FEC / DTX.
• Verify decode of the camera and phone formats people actually bring, including variable frame rate; document what is not supported.

G7. Learned models for Roto Brush and face tracking

• Decided on 5 October 2026 for Roto Brush: licensed weights with pure-Rust inference, if the impact on the system is low, in a composable, swappable module. Done in M13.35: the anima-segment crate (a MaskModel interface, a registry that only accepts open-source licences, MobileSAM under Apache-2.0), weights downloaded on demand and verified, Settings ▸ Roto Brush to choose. Next: measure it against After Effects (G1); faster encoders (GPU) for long shots.
• Face tracking: decided the same way on 5 October 2026, done in M13.36: MediaPipe Face Landmarker (Google, Apache-2.0: BlazeFace detector and the 478-point Face Mesh V2) on a pure-Rust TensorFlow Lite interpreter, behind a FaceModel interface in the same crate, a 3.8 MB optional download chosen in Settings ▸ Face Tracking. The classical fitter stays as the fallback. Next: measure both against After Effects (G1).

G8. Plug-in ecosystem

• After Effects SDK plug-ins cannot run in Anima. Our own WebAssembly plug-in API exists (plugins.md). Grow it: documentation, examples, and original effects that cover what the most common third-party plug-ins are used for (particles, glows, 3D objects, sabers).
• Features that come from third parties in After Effects are extensions in Anima too, built on the same extension points as After Effects: scripts, ScriptUI panels and effect plug-ins (plugins.md compares them). ScriptUI panels in the user's ScriptUI Panels folder are listed in the Window menu, dock like any panel and open again at the next launch. Missing: the Startup and Shutdown script folders, keyboard shortcuts for scripts and .jsxbin; #include / #target directives landed (#759, scripting-parity.md).
• Done so far: ease presets (#254) cover what ease-curve plug-ins are used for: easing curves kept by name and applied to pairs of keyframes. Modelled on a third-party panel, they are not core: a ScriptUI panel bundled as an optional extension (extensions/scriptui-panels/Ease Presets.jsx, Window ▸ Ease Presets.jsx) on the public scripting API, which needed app.settings that last, per-key methods that keep the key selection, and eases reported as they play (plugins.md, target-app-parity.md).

G9. Reach

• Localisation of the interface (localization-parity.md): since 6–12 October the menus are translated into Japanese, Simplified and Traditional Chinese, Ukrainian, Spanish, Brazilian Portuguese, Russian, French, German, Czech and Georgian, and the panel catalog covers Japanese, both Chinese, Ukrainian, Brazilian Portuguese, Czech and Georgian. Next: effect and parameter names, the remaining panel strings (Spanish, Russian, French, German), Korean and Italian, and native-speaker review of all of them.
• Accessibility (keyboard navigation, screen readers, contrast), user documentation and tutorials.

For agents choosing work

1. Read plan/STATUS.md for owner blockers and running work, then this file.
2. Prefer G1–G3 over new features. A feature that exists but behaves differently from After Effects is not done.
3. When you fix a behaviour bug in a feature target-app-parity.md marks as done, add it to the evidence above, and update its row in the ranked list (remove it when the gap is closed). That is how this assessment stays honest.
4. Don't raise target-app-parity.md's numbers without evidence: a G1 score, a user-facing check on more than one platform, or both.

Revision history