Comfyui-EzFlex-Presets

Comfyui-EzFlex-Presets (V1.3.0 Beta3)

English | 中文

A flexible combo plugin for ComfyUI, built with AI; still being improved.


Demo video (Bilibili): watch

EzFlex node list:

• Main Control (MainControl)
• Models Combo Loader (EzFlex-ModelsCombo)
• Resolution / Latent Selector (EzFlex-FreeLatent)
• Node Switch Master (NodeSwitchMaster)
• Node Switch Group (NodeSwitchGroup)
• Param Preset Control (ParamPresetControl)
• Param Preset Output (ParamPresetOutput)
• Preview Any (EzFlex-PreviewAny)
• Prompt Helper (EzFlex-PromptHelper)
• Media Loader (EzFlex-MediaLoader)
• Media Out (EzFlex-MediaOut)
• Merge List (EzFlex-MergeList)
• Split List (EzFlex-SplitList)
• Loop Start (EzFlex-LoopStart)
• Loop End (EzFlex-LoopEnd)
• Time Line (EzFlex-TimeLine)
• Reroute (EzFlex-Reroute)
• 17 nodes in total

Version history:

• V1.3.0Beta3: Optimized the find/replace display in Prompt Assistant, and optimized PreviewAny’s save logic (to avoid overwriting).
• V1.3.0Beta2: Optimized the operation logic of Prompt Assistant, and added a right-click quick tool feature.
• V1.3.0 (earlier in this cycle): Optimized random tags; fixed category selection diverging from the in-library categories; added a duplicate-category warning (a child category duplicating its parent); the card manager gained image generation; reworked MediaLoader / MediaOut port types and interaction; reworked PreviewAny preview & save logic; added loop nodes — Merge List (EzFlex-MergeList), Split List (EzFlex-SplitList), Loop Start (EzFlex-LoopStart), Loop End (EzFlex-LoopEnd), Time Line (EzFlex-TimeLine) — the Reroute node (EzFlex-Reroute), and a loop system for MediaLoader and PromptHelper; tag / card previews are stored as standalone image files (tag\\\_preview / card\\\_preview), with Ctrl+Z / Ctrl+Y guarded so panel undo does not touch the graph.
• V1.2.12: Tag-system performance improvements; fixed cards failing to fill content; fixed the display conflict between the tag hint and "@" media reference; fixed PreviewAny failing to save some media types; improved the save logic.
• V1.2.11: Fixed several PromptHelper frontend bugs and the API-call issue; added API-call error hints; overhauled the tag system: random match/exclude field filters, related tags, a "Hide NSFW" filter, finer in-library tag classification, and a new anima 2.9B tag library.
• V1.2.10: Fixed the ModelsCombo first output port disconnecting after switching windows, refreshing, or restarting; fixed preset selections being lost after switching windows or refreshing.
• V1.2.9: Fixed Preview Any reading incorrect image generation info (prompt text is now read through connected nodes); fixed Node Switch Group possibly failing to match nodes after being copied. ModelsCombo keeps the trigger-words output port whenever it has a config (empty without LoRA).
• V1.2.8: Preview Any (EzFlex-PreviewAny) supports batch image preview; the preview window behavior was improved.
• V1.2.7: Fixed the preview-image disappearing bug, improved black-label text rendering, updated to align with official node features.
• V1.2.6: Added the theme system, optimized the FreeLatent canvas, optimized tag previews, normalized data storage, cleaned up some redundant code, and fixed some bugs.
• V1.2.5: The tag system gained random tags; the default tag library is now bundled (it was missed last time); preview generation gained a cleanup action. ModelsCombo can auto-generate the trigger words of a loaded LoRA (shown live in PromptHelper), and the browser gained a "loaded models" view. Fixed black tag sub-nodes not showing and residue after fast moves. Overall edit supports collapsing individual cards.
• V1.2.4: Prompt Helper big update: new tag system / card manager; improved security.
• V1.2.3: Fixed some bugs, continued comprehensive security improvements, enhanced node features.
• V1.2.2: Fixed the EzFlex-NodeSwitchGroup dropdown bug; fixed the bug where EzFlex-ParamPresetOutput, EzFlex-PreviewAny etc. could not validate input after switching presets in EzFlex-ParamPresetControl; optimized where the custom size presets sit in EzFlex-FreeLatent. Fixed the unclickable controls in the Media Loader preview popup. Fixed Media Loader loading 3D models and preview images too slowly. Fixed PromptHelper's highlight logic.
• V1.2.1: Fixed the Node 2.0 click bugs in Prompt Helper (EzFlex-PromptHelper), Media Loader (EzFlex-MediaLoader) and Media Out (EzFlex-MediaOut). Prompt Helper (EzFlex-PromptHelper) and Node Switch Group (NodeSwitchGroup) gained collapsible rows. Comprehensive security hardening across the plugin.
• V1.2.0: Comprehensive security hardening; Main Control (MainControl) gained an EN / 中文 toggle.
• V1.11: Prompt Helper overhaul and bug fixes.
• V1.1: General node polish and performance work; Prompt Helper improvements; compatibility fixes.
• V1.05: Prompt Helper gained "reference media" and a stronger @media reference.
• V1.04: Added Media Loader (EzFlex-MediaLoader) and Media Out (EzFlex-MediaOut); Prompt Helper overhaul (still in development).
• V1.03: README cleanup; FreeLatent gained a "force override" button (ignore external width/height/batch); fixed the align default (multiples of 8); added the Prompt Helper node.
• V1.02: ModelsCombo gallery/search improvements.
• V1.01: ParamPresetControl preset switching + wiring fix.
• V1.0: First stable release.

Install

• Put this folder under ComfyUI/custom\\\_nodes/ and restart ComfyUI.
• Dependencies: the plugin only needs mutagen as an extra (to read audio/video tags); the rest (torch / numpy / Pillow / safetensors / av) ships with ComfyUI. If you don't want to install it manually, run pip install -r requirements.txt.
• Optional (deliberately not in requirements, so ComfyUI-Manager never forces a compile — install by hand when needed): llama-cpp-python (Prompt Helper's "in-process llama-cpp-python" mode), gguf / onnx (model metadata cards for those two formats). Or install everything at once: pip install -e .\\\[llama,metadata].

Usage

Search EzFlex in the ComfyUI node list and pick the node you need.

Node features

Main Control (MainControl):

• Control: drives the overall node behavior; presets can be freely combined, saved and deleted.
• Card list: Models Combo Loader (EzFlex-ModelsCombo), Resolution / Latent Selector (EzFlex-FreeLatent), Node Switch Master (NodeSwitchMaster) and Param Preset Control (ParamPresetControl) — each with its own preset dropdown and drag-to-reorder. A master preset records their current presets together and cascades them down when switched. Node Switch Group is not listed here separately — it is managed by Node Switch Master.
• Quick load: load the other EzFlex nodes in one click (Media Out (MediaOut) is loaded by Media Loader (EzFlex-MediaLoader) itself).
• Language: switch between 中文 / English.
• Theme: one palette switch recolors every EzFlex panel (Light / Lilac / Sage / Nord / Minimal Cool Grey / Warm Caramel / Cool Haze Blue / Deep Space / Morandi Purple-Grey / Mermaid Core / Chocolate / Klein Blue / Cloud White / Banana Yellow / Burgundy / Deep Teal — 16 in total). Palettes live in the theme module and the panels only write CSS variables, so a switch repaints all open panels immediately.

Models Combo Loader (EzFlex-ModelsCombo):

• Combo loading: freely combine UNet / CLIP / VAE / Checkpoint / LoRA loaders.
• Presets: freely combine / save / delete model-loading setups.
• Preview: preview dropdown-list model covers.
• Gallery: gallery-style browsing to pick models. (Needs the ComfyUI-Lora-Manager plugin to generate JSON files via Civitai.)
• Outputs: ports are generated by the number (loader cards) and type; the LoRA loader chains after its selected target model, multiple LoRAs chain in card order, and port names look like custom\\\_name\\\_model/clip/vae.
• Order: reorder cards freely.

Resolution / Latent Selector (EzFlex-FreeLatent):

• Canvas: drag freely to generate the matching empty latent; hold Ctrl to disable snapping.
• Width/height input: type width/height manually.
• Align resolution: computes width/height with the Opt / Std algorithm and the resolution step; it currently affects manual width/height input, size-preset selection, and sizes derived from ratio and MP.
• MP: megapixels.
• Ratio presets: width : height; choose built-in and custom ratio presets.
• Size presets: choose built-in and custom size presets; custom ones can be saved/deleted (based on the current actual size).
• Custom ratios: save / delete your own input ratio presets.
• Info panel: live actual width/height, ratio and MP.
• Inputs: external width/height, batch size.
• Outputs: empty latent, width/height, batch size.
• Force override: when green, external width/height/batch are ignored and the panel values win, and the controls are unlocked (available only when at least one input has a value).

Node Switch Master (NodeSwitchMaster):

• Control: drives Node Switch Group behavior; freely combine / save / delete presets.

Node Switch Group (NodeSwitchGroup):

• Control: switch grouped nodes between on / off / bypass (ignore).
• Presets: freely save / delete node switch presets.
• Group matching: by name / by color, subgraph matching.
• Order: auto-sort cards by position / name.
• Note: presets are per node instance; deleting the node removes its presets.

Param Preset Control (ParamPresetControl):

• Control: green/red parameter buttons: whether a parameter shows in the output list and in the parameter-group card dropdown. Parameter group dropdown: select the output parameter (all or a specific one).
• Presets: freely save / delete parameter-group presets.
• Parameters: eight types — int, float, bool, string, complex, tuple, list, set, dictionary.
• Order: parameter-group cards and parameter cards can both be dragged freely.
• Outputs: one output port per parameter-group card; multiple parameters are red, a single parameter is gray.

Param Preset Output (ParamPresetOutput):

• Control: enable/disable parameter output; int->0, bool->false, float->0.0, string etc.->empty string.

• Type hint: shows the actual type; if the value does not match the type set in Param Preset Control (ParamPresetControl), it outputs string by default and is marked red.

• Value preview: click to open a preview popup.

Preview Any (EzFlex-PreviewAny):

• Preview: auto-detects any input type and renders the matching preview card (drag to reorder, cards are added/removed with connections).
• Save: preview-only by default; click "auto save" to write to disk. Save options follow the format (lossless formats take no bitrate, lossless images take no quality) and IMAGE / AUDIO / VIDEO offer "Keep source" (copy the source file as-is, no re-encode). File names support %year% %month% %day% %hour% %minute% %second%.
• Batch: one card exports at most 64 frames; the multi-image save mode (auto / batch images / video / animation) is chosen per card.
• Supported types, preview and saving details:

|Type|Accepts|Preview|Formats| |-|-|-|-| |IMAGE|tensor \\\[B,H,W,C]|thumbnail + full-screen original (batches: thumbnail strip paging; saves as image sequence / video / animation)|PNG / JPEG / WebP / BMP / TIFF / animation (WebP / PNG / GIF)| |MASK|2D/3D tensor|grayscale PNG|PNG| |AUDIO|{waveform,sample\\\_rate} or (waveform,sr) or a file object|player|WAV / MP3 / FLAC / OGG / M4A / AAC| |VIDEO|VideoFromFile / VideoFromComponents object|first-frame cover + player|MP4 / WebM / MOV / GIF / AVI / MKV| |CONDITIONING|a dict with conditioning/context|text summary|—| |LIST / TUPLE / SET|list / tuple / set|indexed value tree (index: value)|JSON| |DICT|dict|key/value tree (key: value)|JSON| |STRING / INT / FLOAT / BOOLEAN|str / int / float / bool|text (truncated + popup full text)|TXT / MD / JSON / CSV / LOG / HTML| |LATENT|{samples} dict|shape / dtype summary|—| |FILE_3D|File3D object (same as the built-in Load3D)|three.js viewer (offline)|glb / gltf / obj / fbx / splat| |MODEL_3D|File3D object (with path/file)|three.js viewer (offline)|glb / gltf / obj / fbx / splat| |MESH|Types.MESH (vertex/face tensors, Hunyuan3D / Trellis / MoGe …)|vertices/faces exported to a temp OBJ → three.js viewer|vertices ≤500k, faces ≤1M (beyond that: summary only)| |SPLAT / VOXEL|Types.SPLAT / Types.VOXEL (tensors)|text summary (point count / SH coefficients / voxel resolution)|—| |MODEL|ModelPatcher|model metadata card|safetensors / gguf / onnx / ckpt / pt| |CLIP|comfy.sd CLIP|metadata card|safetensors / gguf / onnx| |VAE|comfy.sd VAE|metadata card|safetensors / gguf / onnx| |CONTROL_NET / CLIP_VISION / STYLE_MODEL / UPSCALE_MODEL / LORA_MODEL / GLIGEN / SAMPLER / SIGMAS / GUIDER / NOISE / SEGS|the matching ComfyUI object|text summary|—| |EMPTY|not connected|"(not connected)"|—|

• Saving details: preview-only by default; click "auto save" to write to disk, with an optional save location and save options.
• Save options follow the format: lossless formats take no bitrate, lossless images take no quality; IMAGE / AUDIO / VIDEO offer "Keep source" (a file loaded from disk is copied as-is, no re-encode).
• File name: supports %year% %month% %day% %hour% %minute% %second%; if left empty the card name is used with an automatic number.
• Model types (MODEL / CLIP / VAE / LORA_MODEL): only the info text is saved (summary + metadata) — the model file is not copied.
• Batch images: one card exports at most 64 frames (beyond that only the count is recorded); the corner badge shows images / frames by the upstream node class name (video/frame/sequence …) — batch image output and video frame extraction cannot be told apart on the tensor.
• Multi-image save mode is chosen per card ("multi-image save type": auto / batch images / video / animation, default auto, only for multi-image cards): auto = extract frames → video, batch output → one by one; the animation format / fps / lossless are set under "save type". File name is name\\\_index\\\_NN. Non-multi-image cards are saved normally by type.
• Preview: data preview popup / generation info.

Prompt Helper (EzFlex-PromptHelper):

• Cards: add / remove / drag to reorder / double-click to rename; each card has a "Default / Optimized" pair of pages and a "merge" toggle (green = merged, gray = not merged).

• Editor: Word-style toolbar (bold/italic/underline/strike, alignment, font size, font color, highlight, first-line indent, find & replace, color picker, full/half-width conversion) plus a skill button (pick a *.md and insert it line by line at the cursor).

• Reference media: reads the media ports of generation nodes on the canvas live and numbers them as @image N / @video N / @audio N; the "reference media" window lists material per generation node, with + insert / − remove / right-click to set as the global reference library.

• Prompt rules: 15 built-in rules under "Settings · Rules" (including "no compile" and "use API") plus custom ones (editable, savable, with 中/EN variants), compiling reference marks into each vendor's syntax (<Picture 1>, @image1, …).

• Optimize: the tool dropdown offers "optimize prompt (API) / (TextGenerate) / (llama)"; use it once or enable "optimize at runtime". The three toggles are mutually exclusive and failures are reported directly.

• Quick weight: with the caret on a tag, press Ctrl+↑ / Ctrl+↓ to add/subtract the "weight step" (Settings · Other settings, default 0.05) — e.g. long_hair → (long_hair:1.05), then back to long_hair once the weight returns to exactly 1 (the brackets are dropped). It follows the caret, so moving with the arrow keys first weights whatever tag the caret is now on. Clamped to 0.05 – 10.

• Collapse: two hover-only bars in the card popup / overall edit fold the toolbar and the Default / Optimized tab row separately (the text area is never folded). Collapse state is saved in the node config and survives a refresh; when both are collapsed they merge into a single "expand all" bar that is again hover-only.

• Tag entry: a Tag button sits to the right of Hint in the card popup / overall edit footer, opening the tag panel — it stays available even when the prompt tools are collapsed.

• Card manager: store cards as presets (user/EzFlex/prompts/<name>.json); save only the checked ones, click to load, delete.

• Ports: no CLIP input — a dynamic "combined media" port (ANY: images / video / audio / 3D) plus one text input per card; outputs are "merged prompt" plus one per card.

• Settings: six tabs — General / Rules / API / TextGenerate / llama / Paths.

• Mode: the header's "⧉ tile / 🗗 popup" switches the window shape in one click.

• Optimization rules & rule compilation:

  • Two sliders: the "Default / Optimized" pages in a card popup only affect that card's output port; the ones in overall edit affect the "merged prompt" port. Each layer's optimized content is its own slot — they do not overwrite each other.
  • Tool optimization — card popup "Tool → optimize prompt": the source is fixed to that card's Default page; the result is written to that card's Optimized slot and the slider switches to Optimized, the Default body is untouched. Overall edit "Tool → optimize prompt": the source is each card's Default body joined in order with the "card merge separator" (merged = green card; empty cards take no slot), one single call; the result goes to the "overall optimization result" and the overall-edit slider switches to Optimized, the original cards are unchanged. Both obey the same table:

|Slider|Default|Optimized|Behavior| |-|-|-|-| |Default / Optimized|has|none|optimize using the default prompt → fills the optimized slot, slider switches to Optimized| |Default / Optimized|has|has|optimize using the default prompt → overwrites the optimized slot (does not reuse hand-edited optimized text)| |any|empty|has / none|pops "this card has no prompt content to optimize."|

• Runtime optimization — any of the three auto-optimize toggles on: on failure (API / TextGenerate / llama error) it stops with red. The optimized slot is a "memo": if it has content it is not re-called, only when empty. Overall layer → "merged prompt" (source = the merged Default bodies of merged=green cards, joined with the card merge separator):

|Overall slider|Default|Optimized|Behavior| |-|-|-|-| |any|has|none|optimize once overall → write to "overall optimization result" → slider switches to Optimized → output it| |Optimized|has/empty|has|output the "overall optimization result" directly (no re-optimize), slider unchanged| |Default|has|has|output the default merged (the optimized version is kept unused, and no optimization), slider unchanged| |Default|empty|has|output the "overall optimization result", slider switches to Optimized| |any|empty|empty|output empty (the user wrote nothing), slider unchanged|

• Each card → "card i" port: when merge is green (joins the merge) it is not optimized separately at runtime, strictly output by that card's slider — Default outputs Default, Optimized outputs Optimized; pointing at an empty slot outputs empty (no fallback). The card list tag follows the slider ("opt" / "def"). When merge is gray (does not join the merge) it is optimized separately on demand at runtime (the result only reaches its own port):

|Slider|Default|Optimized|Card port| |-|-|-|-| |any|has|none|optimize the default separately once → fills that card's optimized slot → port = result, slider switches to Optimized| |Default|has|has|port = Default (the optimized version is kept unused, and no optimization), slider unchanged| |Optimized|has|has|port = optimized version (no re-optimize), slider unchanged| |any|empty|has|port = optimized version (no re-optimize); if the slider was on Default it switches to Optimized| |any|empty|empty|port = empty, slider unchanged|

• Port count = number of cards + 1 (port 0 is "merged prompt", then "card 1..N" all STRING); ports follow card add/remove/sort. If card\\\_in\\\_i is wired to external text it overrides that card (the card is dimmed on the panel) and also takes part in the merge and optimization.
• All three toggles off: no optimization call is made, with the same "strictly by slider, empty stays empty" rule. Overall edit: slider on Default → outputs the default merge (each card's Default body, filtered by merge, gray cards excluded); on Optimized → outputs the "overall optimization result". Empty stays empty, no fallback, no filling in. Each card (green and gray the same): Default → that card's Default body; Optimized → that card's optimized content; empty stays empty. A card slider does not affect the "merged prompt" — the merged port only takes each card's Default body (this is why a card's optimized content only goes through "its own port").
• Rule compilation (referencing media), and at which step: the body sent to the API / TextGenerate / llama is the uncompiled original text (the @image1 / @video1 / @audio1 you typed); the system prompt explicitly asks the model to keep those marks as-is (no translating, rewriting or renumbering). Both optimized slots (that card's "optimized prompt" / "overall optimization result") also store the model's original text (uncompiled) — you can still recompile with a different rule set and never lose the mapping. Only the output ports go through rule compilation: the "card i" port, and the default merge / overall-optimized content used by the "merged prompt"; compilation is idempotent. If the model keeps @image1 → the output automatically becomes that rule's syntax (e.g. <Picture 1>); if the model translates/rewrites it into natural language ("the first picture"), the mark is gone and no compilation can recover it (only the system prompt can lower the odds, or use a more obedient model).

Media Loader (EzFlex-MediaLoader):

• Presets: save and delete group and card-group state plus names.
• Parameters: cards per row, card height multiplier (default 1).
• Groups / card groups / media cards: drag to reorder, double-click to rename, context menu, long-press drag to merge, hover info, click to preview, delete.
• Browse: file-explorer style (directory tree, back/forward/up/refresh, manual path entry, search, multi-select).
• Output: one port per card, of the dedicated EZFLEX\\\_MEDIA\\\_CARD type (dark red) — it carries the card object, not a media value, so it can only connect to EzFlex-MediaOut (this blocks accidental wiring into built-in nodes); for real media values use MediaOut.
• Loadable types: images / video / audio / 3D models / text / other, with the same output values as ComfyUI's built-in loaders, so they wire straight into standard nodes.
• Supported extensions: images .png .jpg .jpeg .webp .gif .bmp .tif .tiff | video .mp4 .webm .mov .mkv .avi .m4v | audio .mp3 .wav .flac .ogg .aac .m4a .opus .wma | 3D .obj .glb .gltf .fbx .stl .ply .spz .splat .ksplat .3ds .dae .blend | text/other → STRING.
• Top bar "load output": creates one EzFlex-MediaOut and wires it for you.
• ⚠️ Be careful with the root-directory (browsable roots) setting: by default you can only browse ComfyUI's own input / output; to reach other directories, type a path in the toolbar and click "save root" to register it explicitly. Only people who can reach your ComfyUI port can see those roots — when it listens on 127.0.0.1 (the default) adding roots only affects you; but once you use --listen 0.0.0.0, a tunnel or a reverse proxy to the internet, anyone on that network/internet can list and download the directories you registered. So: never register a drive letter, your home directory or a project root (the plugin also refuses whole drives and clears legacy ones); registering/deleting is local-only; and delete roots when you are done (the built-in input / output cannot be deleted).

Media Out (EzFlex-MediaOut):

• Input: a single input (dedicated type EZFLEX\\\_MEDIA\\\_CARD), wired to one card port of EzFlex-MediaLoader.
• Output: neutral ports colored/typed by the real file type — IMAGE / VIDEO / AUDIO / FILE_3D / STRING (same as the built-in loaders, so they wire straight into standard nodes).
• Modes: split / card / card group / group (switched from the panel header); split → one port per file, card / card group / group → merged into one port following the structure.
• Merge: multiple images → a batch tensor \\\[B,H,W,C] (accepts any IMAGE input, and can be split again with the built-in ImageFromBatch / RebatchImages); multiple audio files → concatenated into one track (sample rates must match; mono is upmixed); multiple texts → joined with newlines; mixed types / mismatched sample rates → an error suggesting "split" (the port degrades to \\\* and the hover label shows ×N).
• Batch settings (shown in card / card group / group mode, aligned with KJNodes Load Images From Folder): target size use first / specify size, fit crop / pad / stretch, max items (0 = all), start index (0-based).
• Switches: each can be toggled; in split mode a disabled port is kept and outputs None (re-enabling needs no re-wiring), in the other modes a disabled item is dropped from the group. At runtime (before the prompt is queued) an input that is "disabled but still connected" is removed from the submitted prompt — downstream treats it as not connected, and neither the canvas wiring nor the saved workflow changes.

Merge List (EzFlex-MergeList):

• What it does: joins multiple inputs, in order, into one list.
• Input ports are added/removed as needed (input\\\_1..20); a list input is flattened in, anything else is appended as one item.
• Outputs one \\\* list port, ready for EzFlex-SplitList or another loop node.

Split List (EzFlex-SplitList):

• What it does: splits one list into multiple output ports (the inverse of Merge List).
• Output ports are added/removed as needed, always keeping at least one; unconnected ports get no black label.
• Commonly used to take the segments of a collected loop list out one by one.

Loop Start (EzFlex-LoopStart):

• What it does: the loop entry point. The panel has a single index field — which round this run starts from (0 by default). At runtime it is read-only and increments every round.
• Outputs: index (INT, always first — wire it into PromptHelper / MediaLoader / TimeLine's index input) plus value1, value2, … (dynamic, added one at a time as you connect).
• Inputs: index (optional — feed an upstream value when you want to drive the start round from the graph) plus initial value1, initial value2, … (dynamic, one per value port). This is the first-round seed: until feed-back arrives, valueN outputs the matching initial valueN (left unwired, it outputs nothing). Wire the same kind of data you expect Loop End port N to carry back — you need it whenever a downstream node cannot accept a missing value (torch.cat / VAE / …).
• valueN is the fed-back value: whatever Loop End's port N received last round. From round 2 on it overrides the initial value.
• The round count does not live here — it is Loop End's panel field.

Loop End (EzFlex-LoopEnd):

• What it does: the loop trigger and exit point. The panel has a single round counter, and it counts total rounds (the first one included): 0 or 1 = a single pass (a plain run), N = N passes total.
• Inputs value1, value2, … (dynamic — added one at a time as you connect, one-to-one with Loop Start's outputs): whatever a port receives becomes the next round's Loop Start value N.
• Once the total round count is reached, these values are returned as-is on out1, out2, … (the values of the last round — wire a EzFlex-MergeList output into valueN if you want outN to be the accumulated per-round list).
• ⚠️ Do not wire Loop End back into Loop Start on the canvas — the feed is automatic, and a real wire closes the graph cycle (Dependency cycle detected).
• ⚠️ outN can only feed nodes outside the loop body: the body is everything downstream of Loop Start (up to Loop End), so LoopEnd.out1 → TimeLine.video is a cycle when TimeLine also takes LoopStart.index. Use LoopStart.valueN → TimeLine.video instead (valueN is the previous rounds' accumulated value).

Time Line (EzFlex-TimeLine):

• What it does: slices a total duration into segments on the model's frame grid, giving each segment's frame count and the total segment count.
• Inputs: index (from the loop's index) and video (segment video list, item i = segment i).
• Outputs: frames (frames to generate for this segment), overlap\\\_frames (overlap frames), segments (total segments; wire into LoopEnd's round counter when the round count is not known up front).
• Ships with minimax_h3 / wan / ltx model parameters; the segmentation is visualized in the "big preview on top + splice track at the bottom" panel.

Reroute (EzFlex-Reroute):

• Relay: a mid-flow reroute node for larger workflows.
• Connections: adds connectable ports as you wire, and each relay card can be named freely.
• Colors: port and link colors follow the upstream output port (VAE red / CLIP yellow / MODEL purple …), so you can see at a glance what is being relayed.

Directory structure

Comfyui-EzFlex-Presets/
├── \\\_\\\_init\\\_\\\_.py          # all 17 node classes + preset routes + output-type sync routes
├── pyproject.toml
├── README.md            # English README (default home page)
├── README\\\_ZH.md         # Chinese README
├── docs/                # detailed references and research notes (loop mechanics / long video / director …)
├── user\\\_data/           # preset library (written at runtime by the plugin)
├── locales/zh/nodeDefs.json  # official i18n: Chinese node names / descriptions / tooltips
└── web/                 # frontend panels (one \\\*\\\_node.js per node; shared ezflex\\\_service/i18n/theme)
    ├── ezflex\\\_service.js       # shared: registry / group matching / preset API / panel helpers
    ├── ezflex\\\_i18n.js          # panel i18n: ezT(key) + Chinese dictionary (English is the source)
    ├── ezflex\\\_theme.js         # theme: 16 palettes → shared CSS variables (--ez-\\\*)
    ├── ezflex\\\_media\\\_index.js   # media numbering engine (generator media ports → image N / video N / audio N)
    ├── ezflex\\\_listview.js      # list / text preview popup
    └── libs/ utils/ curves/    # three.js and loader/curve resources (local offline, for the 3D viewer)

Notes

• A personal project, built with AI assistance and still being improved — feedback and suggestions are welcome.
• Presets and caches live under user/EzFlex/; workflows only store each node's own configuration.