Upload 2 files

app.py

@@ -1616,7 +1616,8 @@ def delete_profile(name):
 # ----------------------------------------------------------------------------------------------------------------
 
 
-def gpu_estimate(canvas, duration, steps, match, audio_path, video_path, identity_ref, *rest):
+def gpu_estimate(canvas, duration, steps, match, audio_path, video_path, identity_ref,
+                 clip_count=1, *rest):
     """What this request will reserve, by the same `budget()` the pre-flight check and `get_duration` use.
     The identity face counts too: it is sent as one extra picture, so it adds its rows."""
     images = list(rest[:MAX_IMAGE_SLOTS])
@@ -1661,7 +1662,9 @@ def gpu_estimate(canvas, duration, steps, match, audio_path, video_path, identit
         advice = f"Lower Steps to {room}." if room >= MIN_STEPS else "Lower the duration or pick a smaller canvas."
         head = f"🚫 **Wants ~{int(total)} s of GPU, ceiling is {MAX_GPU_DURATION} s.** {advice}"
     else:
-        head = f"⏳ **GPU cost: ~{reserved} s**"
+        clips = max(1, int(clip_count or 1))
+        head = (f"⏳ **GPU cost: ~{reserved} s** for one clip  \n"
+                f"🎬 whole scene, {clips} clips: **~{reserved * clips} s**")
 
     detail = (
         f"{width}x{height} · {seconds:.1f} s ({num_frames} frames) · {int(steps)} steps · "
@@ -1670,6 +1673,7 @@ def gpu_estimate(canvas, duration, steps, match, audio_path, video_path, identit
     )
     if derivable and match:
         detail += " · duration comes from the reference soundtrack"
+    detail += " · joining the clips runs on the processor and costs nothing"
     return f"{head}  \n<sub>{detail}</sub>"
 
 
@@ -1821,6 +1825,63 @@ CIVITAI_API_HOSTS = [h for h in (os.environ.get("CIVITAI_API_HOST", "").strip(),
 H3_BASE_MODELS = ["MiniMax H3", "(any base model)"]
 
 
+_SCENE_INDEX = {}
+
+
+def _scene_prompt_lines(text, total, fallback):
+    """One prompt per clip. A line the user left empty - or a line that is not there at all -
+    means that clip runs on the main prompt, so the box can be left alone entirely."""
+    lines = str(text or "").splitlines()
+    out = []
+    for index in range(total):
+        line = lines[index].strip() if index < len(lines) else ""
+        out.append(line or fallback)
+    return out
+
+
+def make_scene(clip_count, scene_prompts, randomize, queue, name_hint, *request_values):
+    """One button, a whole scene. Generates `clip_count` clips back to back - each one starting on
+    the last frame of the clip before it - and stitches them into a single video as it goes, with
+    the soundtrack carried through. The identity face stays locked across the whole run, so the
+    person does not drift from clip to clip."""
+    values = list(request_values)
+    idx_prompt = _SCENE_INDEX.get("prompt", 0)
+    idx_image = _SCENE_INDEX.get("image", 1)
+    idx_seed = _SCENE_INDEX.get("seed", 16)
+    idx_identity = _SCENE_INDEX.get("identity", len(values) - 1)
+
+    if not values[idx_image]:
+        raise gr.Error("Add a first reference image before making a scene.")
+
+    queue = list(queue or [])
+    total = max(1, min(8, int(clip_count or 1)))
+    prompts = _scene_prompt_lines(scene_prompts, total, values[idx_prompt])
+    video_path = None
+    refined = ""
+    panel = gr.update()
+
+    # The face that is in view now is what every later clip is pulled back toward.
+    if not values[idx_identity]:
+        values[idx_identity] = values[idx_image]
+
+    for index in range(total):
+        values[idx_prompt] = prompts[index]
+        if index:
+            values[idx_image] = last_frame_of(video_path)
+        values[idx_seed] = roll_seed(randomize, values[idx_seed])
+
+        video_path, refined, panel = generate_with_identity(*values)
+        if not video_path:
+            raise gr.Error(f"Clip {index + 1} produced nothing.")
+
+        if not queue or queue[-1] != video_path:
+            queue.append(video_path)
+        merged = concat_videos(queue, name_hint) if len(queue) > 1 else queue[0]
+        note = (_queue_status(len(queue), merged) if index + 1 == total
+                else f"Clip {index + 1} of {total} done - {len(queue)} in the joined file.")
+        yield (video_path, refined, panel, merged, merged, queue, note, values[idx_seed])
+
+
 def civitai_search(query, base_model, want_nsfw, limit=20):
     """Search CivitAI for lora. `/api/v1/models` embeds each model's versions, files and trigger words, so one call
     gives everything a slot needs. Returns `(readable list, dropdown update, {label: url})`."""
@@ -2524,7 +2585,8 @@ with gr.Blocks(title="MiniMax-H3 - Custom lora + CivitAI, structured prompts, GP
                                                        elem_id="search-put")
                         # Below the picker on purpose: a wall of results above it would push the
                         # controls off the screen, which is exactly what happened the first time.
-                        search_results = gr.Markdown("No search yet.")
+                        with gr.Accordion("📋 The results", open=False):
+                            search_results = gr.Markdown("No search yet.")
                         search_state = gr.State({})
 
                     lora_references, lora_scales = [], []
@@ -2548,7 +2610,8 @@ with gr.Blocks(title="MiniMax-H3 - Custom lora + CivitAI, structured prompts, GP
                     with gr.Row():
                         lora_identify_btn = gr.Button("🔎 name the links", variant="secondary",
                                                       elem_id="lora-identify")
-                    lora_names = gr.Markdown("Nothing in the slots yet.", elem_classes="turbo-blurb")
+                    with gr.Accordion("📋 The named links", open=False):
+                        lora_names = gr.Markdown("Nothing in the slots yet.", elem_classes="turbo-blurb")
                     lora_upload = gr.File(
                         label="Drop .safetensors here to fill the slots",
                         file_count="multiple",
@@ -2613,19 +2676,40 @@ with gr.Blocks(title="MiniMax-H3 - Custom lora + CivitAI, structured prompts, GP
                                              elem_id="estimate-btn")
             with gr.Group(elem_classes="panel"):
                 result = gr.Video(label="🎞️ Video + soundtrack", height=560)
-                extend_btn = gr.Button("➕ Continue the scene (one more clip, joined on)",
-                                       variant="secondary", elem_id="extend-btn")
+                gr.Markdown("### 🎬 Make it longer")
+                chain_count = gr.Slider(
+                    2, 8, value=3, step=1, label="How many clips in a row",
+                    info="3 clips of 5 s ≈ a 15 second video.",
+                )
+                with gr.Accordion("✍️ A prompt per clip (optional)", open=False):
+                    scene_prompts = gr.Textbox(
+                        label="One line per clip", lines=8, max_lines=8,
+                        placeholder=("line 1 = clip 1, line 2 = clip 2, and so on\n"
+                                     "she turns toward the window\n"
+                                     "she smiles and looks down\n"
+                                     "…"),
+                        info="Leave a line empty — or the whole box — and that clip uses the "
+                             "main prompt, unchanged.",
+                    )
+                with gr.Row():
+                    chain_btn = gr.Button("🎬 Make the whole scene and join it",
+                                          variant="primary", elem_id="extend-btn")
+                    chain_stop_btn = gr.Button("⏹ Stop", variant="stop")
+                extend_btn = gr.Button("➕ Just one more clip", variant="secondary")
                 gr.Markdown(
-                    "Takes the last frame of the clip above, makes it the first reference, generates again with "
-                    "the settings untouched, and joins the two into one file — soundtrack included. Press it again "
-                    "for a third clip. Each press costs one normal generation; the joining is free. "
-                    "**The 🔒 identity face stays locked**, so the person never drifts between clips.",
+                    "Press 🎬 once and leave it. Each clip starts on the last frame of the one "
+                    "before it and the joined video below grows as each clip lands — soundtrack "
+                    "included. Every clip costs one normal generation; the joining is free, and "
+                    "**the 🔒 identity face stays locked**, so the person never drifts between "
+                    "clips.  \n"
+                    "➕ does the same thing one clip at a time, for when you want to change "
+                    "something in between.",
                     elem_classes="turbo-blurb",
                 )
             # An output, so it can be revealed only for a request that asked for a rewrite.
             with gr.Accordion("Upsampled prompt", open=False, visible=False) as upsampled_panel:
                 upsampled = gr.Textbox(show_label=False, lines=8, interactive=False)
-            with gr.Accordion("🎞️ Stitch clips into one video (no GPU used)", open=False):
+            with gr.Accordion("🎞️ The joined video (no GPU used)", open=True):
                 gr.Markdown(
                     "Add clips to a queue and they are joined into a single file, **soundtrack "
                     "included**. Three clips make one long video for no extra GPU time. Everything "
@@ -2635,8 +2719,8 @@ with gr.Blocks(title="MiniMax-H3 - Custom lora + CivitAI, structured prompts, GP
                 merge_name = gr.Textbox(label="File name (optional)", placeholder="my_scene",
                                         max_lines=1)
                 auto_merge = gr.Checkbox(
-                    value=False, label="⚡ Add every new generation automatically",
-                    info="Generate, generate, generate — the joined file grows on its own.",
+                    value=True, label="⚡ Add every new clip automatically",
+                    info="Leave this on. 🎬 and ➕ always join their clips whatever this says.",
                 )
                 with gr.Row():
                     merge_add_btn = gr.Button("➕ Add the current video", variant="secondary")
@@ -2722,11 +2806,30 @@ with gr.Blocks(title="MiniMax-H3 - Custom lora + CivitAI, structured prompts, GP
     ).then(
         generate_with_identity, request, [result, upsampled, upsampled_panel], api_name=False,
     ).then(
-        auto_queue, [result, auto_merge, merge_queue, merge_name],
+        add_to_queue, [result, merge_queue, merge_name],
         [merged_video, merged_file, merge_queue, merge_status],
         show_progress="hidden", api_name=False,
     )
 
+    # One button, a whole scene. `make_scene` gets `request` as a flat tuple, so it is told here -
+    # from the list itself - which slots hold the prompt, the first image, the seed and the
+    # identity face, rather than counting positions by hand.
+    _SCENE_INDEX.update(
+        prompt=request.index(prompt),
+        image=request.index(images[0]),
+        seed=request.index(seed),
+        identity=request.index(identity_ref),
+    )
+    _chain_event = chain_btn.click(
+        fn=make_scene,
+        inputs=[chain_count, scene_prompts, randomize_seed, merge_queue, merge_name, *request],
+        outputs=[result, upsampled, upsampled_panel, merged_video, merged_file,
+                 merge_queue, merge_status, seed],
+        api_name=False,
+    )
+    chain_stop_btn.click(fn=None, inputs=None, outputs=None, cancels=[_chain_event],
+                         api_name=False)
+
     # Searching CivitAI, and dropping a result into a slot.
     search_btn.click(
         civitai_search, [search_query, search_base, search_nsfw],
@@ -2804,7 +2907,8 @@ with gr.Blocks(title="MiniMax-H3 - Custom lora + CivitAI, structured prompts, GP
     for _field in lora_references:
         _field.submit(identify_loras, lora_references, lora_names, api_name=False)
 
-    estimate_inputs = [canvas, duration, steps, match, audio, video, identity_ref, *images, *lora_references]
+    estimate_inputs = [canvas, duration, steps, match, audio, video, identity_ref,
+                       chain_count, *images, *lora_references]
     # Recalculated on demand only. An automatic readout means an event on every slider
     # move, and a browser that never gets to rest.
     estimate_btn.click(gpu_estimate, estimate_inputs, estimate, show_progress="hidden")

lora_library.py

@@ -723,16 +723,21 @@ def library_tab(lora_box=None, prompt_box=None, lora_slots=None, scale_slots=Non
         normal_labels = gr.State(_labels(library["normal"]))
         nsfw_labels = gr.State(_labels(library["nsfw"]))
 
-        normal_pick = gr.CheckboxGroup(
-            choices=_labels(library["normal"]), value=[],
-            label="ordinary", interactive=True,
-        )
-        normal_links = gr.Markdown(_links_markdown(library["normal"], "ordinary"))
-        nsfw_pick = gr.CheckboxGroup(
-            choices=_labels(library["nsfw"]), value=[],
-            label="nsfw", interactive=True,
-        )
-        nsfw_links = gr.Markdown(_links_markdown(library["nsfw"], "nsfw"))
+        with gr.Accordion("📋 ordinary", open=False):
+            normal_pick = gr.CheckboxGroup(
+                choices=_labels(library["normal"]), value=[],
+                label="ordinary", show_label=False, interactive=True,
+            )
+            with gr.Accordion("🔗 the links", open=False):
+                normal_links = gr.Markdown(_links_markdown(library["normal"], "ordinary"))
+
+        with gr.Accordion("🔞 nsfw", open=False):
+            nsfw_pick = gr.CheckboxGroup(
+                choices=_labels(library["nsfw"]), value=[],
+                label="nsfw", show_label=False, interactive=True,
+            )
+            with gr.Accordion("🔗 the links", open=False):
+                nsfw_links = gr.Markdown(_links_markdown(library["nsfw"], "nsfw"))
 
         with gr.Row():
             add_button = gr.Button(