FreeVideo 0.4.15

MANIFEST.in

@@ -27,3 +27,5 @@ prune benchmarks
 prune distribution
 prune .github
 exclude AGENTS.md CHANGELOG.md
+
+include freevideo_source_bootstrap.py freevideo_ed25519.py freevideo_update_keys.py

docs/community-results.md

@@ -86,6 +86,10 @@ Thanks to the community members who sent diagnostic reports. FreeVideo already r
 | 4070 Laptop | 8 GB | 34 GB | 1344×768 | 15 s | Light⁵ | 0.4.8 | 1444 s | 0.4.8 | Already optimal¹² |
 | 5060 Ti | 16 GB | 56 GB | 960×544 | 10 s | Light | 0.4.6 | 904 s | 0.4.9 | 805 s (−99 s) |
 | 3070 Laptop | 8 GB | 64 GB | 1344×768 | 10 s | Light | 0.4.9 | 1267 s | 0.4.9 | Already optimal |
+| 3060 | 12 GB | 64 GB | 768×1344 | 5 s | Light | 0.4.12 | 639 s | 0.4.12 | Already optimal |
+| 5080 | 16 GB | 64 GB | 768×1344 | 3 s | Max | 0.3.20 | 216 s | 0.3.20 | Already optimal |
+| 4060 Ti | 8 GB | 32 GB | 960×544 | 5 s | Light | 0.4.12 | 333 s | 0.4.12 | Already optimal |
+| 5070 Ti | 16 GB | 64 GB | 1184×672 | 15 s | Medium | 0.4.13 | 487 s | 0.4.13 | Already optimal |
 
 - Time: end to end, from submission until the video is saved.
 - After: the estimated time, derived from the report's stage times and from complete planned runs on our test machines (RTX 3090 and RTX 5060 Ti on Windows, RTX PRO 6000 on Linux) under the reporting machine's VRAM and RAM limits, not from reruns on that machine.

docs/zh-CN/community-results.md

@@ -86,6 +86,10 @@
 | 4070 Laptop | 8 GB | 34 GB | 1344×768 | 15 s | 轻⁠量⁵ | 0.4.8 | 1444 s | 0.4.8 | 已⁠是⁠最⁠优¹² |
 | 5060 Ti | 16 GB | 56 GB | 960×544 | 10 s | 轻⁠量 | 0.4.6 | 904 s | 0.4.9 | 805 s (−99 s) |
 | 3070 Laptop | 8 GB | 64 GB | 1344×768 | 10 s | 轻⁠量 | 0.4.9 | 1267 s | 0.4.9 | 已⁠是⁠最⁠优 |
+| 3060 | 12 GB | 64 GB | 768×1344 | 5 s | 轻⁠量 | 0.4.12 | 639 s | 0.4.12 | 已⁠是⁠最⁠优 |
+| 5080 | 16 GB | 64 GB | 768×1344 | 3 s | 极⁠致 | 0.3.20 | 216 s | 0.3.20 | 已⁠是⁠最⁠优 |
+| 4060 Ti | 8 GB | 32 GB | 960×544 | 5 s | 轻⁠量 | 0.4.12 | 333 s | 0.4.12 | 已⁠是⁠最⁠优 |
+| 5070 Ti | 16 GB | 64 GB | 1184×672 | 15 s | 标⁠准 | 0.4.13 | 487 s | 0.4.13 | 已⁠是⁠最⁠优 |
 
 - 用时：从提交到视频保存完成的端到端用时。
 - 优化后：预估用时，根据报告中的分段用时，结合测试机（Windows RTX 3090 和 RTX 5060 Ti、Linux RTX PRO 6000）在报告机器的显存和内存上限下运行完整规划的实测推算，不是在报告机器上重跑。

docs/zh-CN/使用说明.md

@@ -47,7 +47,7 @@
 ### 限制
 
 - 两种做法一样，一条视频最多 5 段（第一段加 4 次接着生成），所以创作面板里时长最多填 60 秒。接续的每一段至少 10 秒，整条约 70 秒。剪辑视图里逐段生成同样适用。段数越多，衔接处越容易出现变化，适合切换镜头处，可直接切镜头，用对应的参考图开始新的一段。
-- 接着生成使用轻量档，需要 NVIDIA 显卡，也不能和“继续二采”同时使用。
+- 接着生成使用轻量档，需要 NVIDIA 显卡或 Mac，也不能和“继续二采”同时使用。
 - 单次生成的最长时长随画幅和显卡而不同，在创作面板的“可生成范围”中可查看各尺寸的最长时长和预计用时。
-- 分段生成同样需要 NVIDIA 显卡，不支持 Mac。
+- 分段生成同样需要 NVIDIA 显卡或 Mac。
 - 分段生成过程中取消，只会停止当前这一段，已经完成的段落会保留，之后可以从第一个未完成的段落继续。

freevideo_ed25519.py

@@ -0,0 +1,95 @@
+"""Dependency-free Ed25519 (RFC 8032), with strict encodings/subgroup checks.
+
+Verification handles public inputs. Signing is an offline release utility, not
+constant-time: keep signing seeds on a trusted release machine.
+"""
+import hashlib
+
+P = 2**255 - 19
+L = 2**252 + 27742317777372353535851937790883648493
+D = -121665 * pow(121666, P-2, P) % P
+I = pow(2, (P-1)//4, P)
+IDENTITY = (0, 1, 1, 0)
+
+
+def _decode(data):
+    if len(data) != 32:
+        raise ValueError('Invalid point length')
+    encoded = int.from_bytes(data, 'little')
+    y, sign = encoded & (2**255-1), encoded >> 255
+    if y >= P:
+        raise ValueError('Noncanonical point')
+    x2 = (y*y-1) * pow(D*y*y+1, P-2, P) % P
+    x = pow(x2, (P+3)//8, P)
+    if x*x % P != x2:
+        x = x*I % P
+    if x*x % P != x2 or (x == 0 and sign):
+        raise ValueError('Invalid point')
+    if x & 1 != sign:
+        x = P-x
+    return x, y, 1, x*y % P
+
+
+def _add(a, b):
+    x, y, z, t = a
+    u, v, w, s = b
+    aa, bb = (y-x)*(v-u) % P, (y+x)*(v+u) % P
+    cc, dd = 2*D*t*s % P, 2*z*w % P
+    e, f, g, h = bb-aa, dd-cc, dd+cc, bb+aa
+    return e*f % P, g*h % P, f*g % P, e*h % P
+
+
+def _mul(point, scalar):
+    result = IDENTITY
+    while scalar:
+        if scalar & 1:
+            result = _add(result, point)
+        point = _add(point, point)
+        scalar >>= 1
+    return result
+
+
+def _encode(point):
+    x, y, z, _ = point
+    inverse = pow(z, P-2, P)
+    return ((y*inverse % P) | ((x*inverse % P & 1) << 255)).to_bytes(32, 'little')
+
+
+B = _decode(bytes.fromhex('58' + '66'*31))
+
+
+def verify(public_key, message, signature):
+    try:
+        if len(public_key) != 32 or len(signature) != 64:
+            return False
+        a, r = _decode(public_key), _decode(signature[:32])
+        scalar = int.from_bytes(signature[32:], 'little')
+        neutral = _encode(IDENTITY)
+        if scalar >= L or any(_encode(q) == neutral or _encode(_mul(q, L)) != neutral for q in (a, r)):
+            return False
+        challenge = int.from_bytes(hashlib.sha512(signature[:32]+public_key+message).digest(), 'little') % L
+        return _encode(_mul(B, scalar)) == _encode(_add(r, _mul(a, challenge)))
+    except (ValueError, TypeError, OverflowError):
+        return False
+
+
+def _secret(seed):
+    if len(seed) != 32:
+        raise ValueError('Ed25519 seed must contain 32 bytes')
+    hashed = hashlib.sha512(seed).digest()
+    scalar = (int.from_bytes(hashed[:32], 'little') & (2**254-8)) | 2**254
+    return scalar, hashed[32:]
+
+
+def public_key(seed):
+    scalar, _ = _secret(seed)
+    return _encode(_mul(B, scalar))
+
+
+def sign(seed, message):
+    scalar, prefix = _secret(seed)
+    public = _encode(_mul(B, scalar))
+    nonce = int.from_bytes(hashlib.sha512(prefix+message).digest(), 'little') % L
+    r = _encode(_mul(B, nonce))
+    challenge = int.from_bytes(hashlib.sha512(r+public+message).digest(), 'little') % L
+    return r + ((nonce+challenge*scalar) % L).to_bytes(32, 'little')

freevideo_engine/agent/guide.py

@@ -235,6 +235,10 @@
     'continuing again and again is not recommended. One video has at most 5 parts (the first and 4 '
     'continuations), each continued part at least 10 seconds, about 70 seconds in all; filling that is not '
     'recommended. Past it, cut to a new shot.',
+    'A question is not a request. When the user asks about progress or status ("is it working?", "what happened?") '
+    'answer it in one sentence from what you know or from the job and white-model status, and start nothing; never '
+    'use a question as a prompt. When a row carries "whitebox", the user was starting a white-model scene that is '
+    'still being prepared or failed: say which, and wait for the answer.',
 ]
 
 
@@ -265,6 +269,9 @@
     '默认切镜头。能切镜头的地方就切：新镜头直接用对应的参考图（人物卡、场景图）生成，不接上一段视频。'
     '只有不能切镜的长镜头、必须连贯的动作、必须连贯的声音，才接着上一段生成，而且不建议一段接一段地连续接。'
     '一条视频最多 5 段（第一段加 4 次接着生成），接着生成的每一段至少 10 秒，总长约 70 秒；不建议接满，超过就切镜头。',
+    '问话不是要求。用户问进度或状态（“在做吗”“怎么了”）时，用一句话照实回答，依据是你所知道的、任务和白模的状态，'
+    '什么都不要开始；不要把问句当提示词。某条 inbox 带有 whitebox 时，说明用户刚启动了白模场景，'
+    '它还在准备或已经失败：照实说是哪一种，再等用户回话。',
 ]
 
 
@@ -480,7 +487,7 @@ def connection_guide(root, base, lang=None, offers=None, resumed=False):
                      '不要停下来问用户——你一停，用户在 FreeVideo 里说的话就没人听了。'
                      '用命令行调接口时，每次等待单独发一条命令（一条命令里不写循环，超时设到 60 秒以上），'
                      '返回后接着发下一条。'
-                     '收到话（inbox 里每条的 text）先回一句再动手，例如“好的，开始生成。”：只一句，不复述尺寸、'
+                     '收到要求（inbox 里每条的 text）先回一句再动手，例如“好的，开始生成。”：只一句，不复述尺寸、'
                      '时长这些参数，也不说预计时间（任务卡上都有）。做完再说一句短的，例如“好了。”，'
                      '或者一句用户需要知道的话，发回同一个 session，再接着等。先把话发出去，再调等待接口；不要写成“等待返回后才发话”，那样这句话会晚到将近一分钟。中文用全角标点。'
                      '做完一条视频也不是结尾：这句话就是汇报，不要再在自己窗口里写总结（一写总结这一轮就结束了，'
@@ -498,7 +505,7 @@ def connection_guide(root, base, lang=None, offers=None, resumed=False):
                      'user, because once you stop nobody hears what the user says in FreeVideo. From a command line, '
                      'each wait is its own command (no loop inside one command, timeout above 60 seconds); when it '
                      'returns, run the next one. '
-                     'When words arrive (text of each inbox row), answer with one sentence before you start, for example '
+                     'When a request arrives (text of each inbox row; a question about progress is answered, not started), answer with one sentence before you start, for example '
                      '"OK, starting.": one sentence only, without repeating size or length and without an estimate (the '
                      'job card shows them). When it is done say one short sentence, for example "Done.", or one thing the '
                      'user needs to know, to the same session, and wait again. Post your words first and call the wait after; never wait and then post, or they reach the user up to a minute late. Chinese text takes full-width punctuation. '

freevideo_engine/agent/jobs.py

@@ -1841,6 +1841,9 @@ def start_image(self, row):
         else:
             action = 'generate'
             value.update(prompt=spec['prompt'], images=[image_reference(self.port, name) for name in spec['images']])
+            metadata = spec.get('metadata') or {}
+            if metadata.get('cast_card') and metadata.get('cast_part') == 'back':
+                value['cast_view'] = 'back'
         if row.get('_cancel'):
             raise JobError('cancelled', IMAGE_ERRORS['cancelled'])
         row['_image'] = self.port.image_start(action, value)

freevideo_engine/agent/routes.py

@@ -45,6 +45,10 @@
 RESUMED = ('This conversation is already under way; its last lines are in "before". Do not introduce yourself again: '
            'say nothing until the user speaks, then carry on from there.')
 BEFORE_LINES, BEFORE_TEXT = 8, 600
+# The user started a white-model scene and then spoke: the words may be about it, not a new request.
+WHITEBOX_PENDING = ('The user started a white-model scene in this conversation ("whitebox" lists it). If the words ask about '
+                    'it or about progress, answer from its status in one sentence (waiting or rendering: still being made; '
+                    'failed: say it was not made, and why) and start nothing. A question is never a prompt.')
 
 
 def local_address(request):
@@ -1074,12 +1078,24 @@ def write():
                 raise JobError('not_found', 'No conversation')
             # Work in progress still writes into its conversation; finished videos stay in the library.
             # New work is refused first, so what is counted below cannot grow before the file goes.
+            def busy(leaving=()):
+                return (any(r.get('session') == sid and r['status'] in ACTIVE and r['status'] not in leaving for r in jobs.rows())
+                        or any(f['session'] == sid and not f['paused']
+                               for f in jobs.productions.list(status='active', limit=None)))
+            stop = value.get('stop') is True
+            if stop:  # The user confirmed: stop its jobs and pause its films first, as the stop button does.
+                jobs.stop_session(sid)
+                for _ in range(20):  # A running job takes a moment to reach its cancelled state.
+                    with jobs.lock:
+                        if not busy():
+                            break
+                    time.sleep(0.25)
             store.begin_delete(sid)
             try:
                 with jobs.lock:  # Productions share it.
-                    if (any(r.get('session') == sid and r['status'] in ACTIVE for r in jobs.rows())
-                            or any(f['session'] == sid and not f['paused']
-                                   for f in jobs.productions.list(status='active', limit=None))):
+                    # After the user's stop, a job still winding down no longer holds the conversation: its
+                    # writes are refused once deletion begins, so a stuck one cannot block deleting forever.
+                    if busy(('cancelling',) if stop else ()):
                         raise JobError('conflict', 'Work is in progress in this conversation')
                     store.delete(sid)
             except BaseException:
@@ -1725,6 +1741,10 @@ async def inbox(request):
         follow = store is not None and not request.headers.get('X-FreeVideo-Session', '').strip()
         def answer(value):
             told, chat['told'][client] = chat['told'].get(client), value['session']
+            scenes = jobs.whitebox.open_for(value['session']) if value['inbox'] else []
+            if scenes:
+                value['whitebox'] = scenes
+                value['next'] = (WHITEBOX_PENDING + ' ' + value.get('next', '')).strip()
             if follow and told != value['session']:
                 first = value['inbox'][0]['seq'] if value['inbox'] else store.get(value['session'])['last_seq'] + 1
                 if first > 1:

freevideo_engine/agent/schema.py

@@ -437,7 +437,7 @@ def route(summary, body=None, query=None, **kw):
             ('/activity/cancel', 'POST', obj({'id': JOB}, ('id',))), ('/connect', 'GET', None),
             ('/connect/reset', 'POST', EMPTY), ('/chat', 'GET', None), ('/chat/now', 'GET', None),
             ('/chat/video', 'GET', None), ('/chat/list', 'GET', None), ('/chat/new', 'POST', EMPTY),
-            ('/chat/delete', 'POST', obj({'session': SESSION}, ('session',))),
+            ('/chat/delete', 'POST', obj({'session': SESSION, 'stop': {'type': 'boolean'}}, ('session',))),
             ('/chat', 'POST', obj({'text': string(8000), 'id': CHAT_ID, 'session': SESSION, 'whitebox': {'type': 'boolean'}}, ('text', 'id'))),
             ('/chat/confirm', 'POST', obj({'id': string(12, pattern=r'^cf[0-9a-f]{10}$'), 'approve': {'type': 'boolean'},
                                            'session': SESSION}, ('id', 'approve')))):

freevideo_engine/agent/whitebox.py

@@ -127,6 +127,18 @@ def create(self, code, summary, session=None, notice=None):
             self._save(record)
             return self.view(record)
 
+    def open_for(self, session):
+        """The scenes this conversation started that are still being made, or failed without the user being told."""
+        with self.lock:
+            rows = []
+            for record in sorted(self.records.values(), key=lambda r: r['created_at']):
+                self._expire(record)
+                if record.get('session') == session and record.get('notice') and record['status'] != 'ready' \
+                        and self.clock() - record['created_at'] < 1800:
+                    rows.append(dict(id=record['id'], status=record['status'], summary=record['summary'],
+                                     **(dict(error=record['error']) if record['status'] == 'failed' else {})))
+            return rows
+
     def take_notice(self, identity):
         """What the server should tell the user's conversation once the scene is ready; handed out once."""
         with self.lock: