Merge pull request 'add local image gen, refactor local chat gen' (#12) from ollama-gtx into main

Reviewed-on: #12

garfmain.py

@@ -161,13 +161,13 @@ async def garfchat(ctx, *, prompt):
     await ctx.reply(answer)
 
 
-# @garfbot.command(name="pic")
-# async def garfpic(ctx, *, prompt):
-#     logger.info(
-#         f"Image Request - User: {ctx.author.name}, Server: {ctx.guild.name}, Prompt: {prompt}"
-#     )
-#     await ctx.reply(f"`Please wait... image generation queued: {prompt}`")
-#     await garfield.garfpic(ctx, prompt)
+@garfbot.command(name="pic")
+async def garfpic(ctx, *, prompt):
+    logger.info(
+        f"Image Request - User: {ctx.author.name}, Server: {ctx.guild.name}, Prompt: {prompt}"
+    )
+    await ctx.reply(f"`Please wait... image generation queued: {prompt}`")
+    await garfield.garfpic(ctx, prompt)
 
 
 @garfbot.command(name="help")

garfpy/garfai.py

@@ -1,4 +1,6 @@
 import io
+import time
+import uuid
 import openai
 import config
 import aiohttp
@@ -9,96 +11,212 @@ from openai import AsyncOpenAI
 from garfpy import logger
 
 
+INVOKEAI_BASE = config.INVOKEAI_URL
+
+_MODEL_KEY   = "0eb50094-5c9b-431b-ba01-87e145edb849"
+_VAE_KEY     = "dde3627c-8a45-4088-93d1-66c44acbb337"
+_ENCODER_KEY = "7ba22542-4687-4946-a52e-c92f925f4b75"
+
+_MODEL_REF   = {"key": _MODEL_KEY,   "hash": "blake3:c3ee838d71d99497db01fae6f304eafd9e734e935f3b783e968d50febb56be2c", "name": "FLUX.2 Klein 4B (GGUF Q4)",        "base": "flux2", "type": "main"}
+_VAE_REF     = {"key": _VAE_KEY,     "hash": "blake3:531855de70db993d0f6181f82cde27d15411d58b7ffa3b2fdce2b9434c0173c2", "name": "FLUX.2 VAE",                      "base": "flux2", "type": "vae"}
+_ENCODER_REF = {"key": _ENCODER_KEY, "hash": "blake3:af5840e6770dc99f678e69867949c8b9264835915eb82a990e940fa6e4fa6c81", "name": "FLUX.2 Klein Qwen3 4B Encoder", "base": "any",   "type": "qwen3_encoder"}
+
+_POLL_INTERVAL = 2
+_POLL_ATTEMPTS = 60
+_MIN_IMAGE_BYTES = 1000
+
+
+def _node_id(prefix: str) -> str:
+    return f"{prefix}:{uuid.uuid4().hex[:10]}"
+
+
+def _build_graph(prompt: str) -> dict:
+    seed = int(time.time() * 1000) % (2 ** 31)
+
+    p   = _node_id("positive_prompt")
+    ml  = _node_id("flux2_klein_model_loader")
+    te  = _node_id("flux2_klein_text_encoder")
+    dn  = _node_id("flux2_denoise")
+    out = _node_id("canvas_output")
+
+    nodes = {
+        p:   {"id": p,   "is_intermediate": True,  "use_cache": True,  "value": prompt, "type": "string"},
+        ml:  {"id": ml,  "is_intermediate": True,  "use_cache": True,  "type": "flux2_klein_model_loader",
+              "model": _MODEL_REF, "vae_model": _VAE_REF, "qwen3_encoder_model": _ENCODER_REF},
+        te:  {"id": te,  "is_intermediate": True,  "use_cache": True,  "type": "flux2_klein_text_encoder"},
+        dn:  {"id": dn,  "is_intermediate": True,  "use_cache": True,  "type": "flux2_denoise", "seed": seed},
+        out: {"id": out, "is_intermediate": False, "use_cache": False, "type": "flux2_vae_decode"},
+    }
+
+    edges = [
+        {"source": {"node_id": ml,  "field": "qwen3_encoder"},  "destination": {"node_id": te,  "field": "qwen3_encoder"}},
+        {"source": {"node_id": ml,  "field": "max_seq_len"},    "destination": {"node_id": te,  "field": "max_seq_len"}},
+        {"source": {"node_id": p,   "field": "value"},          "destination": {"node_id": te,  "field": "prompt"}},
+        {"source": {"node_id": ml,  "field": "transformer"},    "destination": {"node_id": dn,  "field": "transformer"}},
+        {"source": {"node_id": ml,  "field": "vae"},            "destination": {"node_id": dn,  "field": "vae"}},
+        {"source": {"node_id": te,  "field": "conditioning"},   "destination": {"node_id": dn,  "field": "positive_text_conditioning"}},
+        {"source": {"node_id": ml,  "field": "vae"},            "destination": {"node_id": out, "field": "vae"}},
+        {"source": {"node_id": dn,  "field": "latents"},        "destination": {"node_id": out, "field": "latents"}},
+    ]
+
+    return {"nodes": nodes, "edges": edges}
+
+
+async def _poll_batch(session: aiohttp.ClientSession, base: str, batch_id: str) -> bool:
+    """Poll batch status until completed, failed, or timed out. Returns True on success."""
+    for _ in range(_POLL_ATTEMPTS):
+        await asyncio.sleep(_POLL_INTERVAL)
+        try:
+            async with session.get(f"{base}/api/v1/queue/default/b/{batch_id}/status") as resp:
+                if not resp.ok:
+                    continue
+                s = await resp.json(content_type=None)
+                total     = s.get("total", 0)
+                completed = s.get("completed", 0)
+                failed    = s.get("failed", 0)
+                if total > 0 and failed >= total:
+                    logger.error(f"Batch {batch_id} failed")
+                    return False
+                if total > 0 and completed >= total:
+                    return True
+        except Exception as e:
+            logger.error(f"InvokeAI poll error: {e}")
+    return False
+
+
+async def _get_image_name(session: aiohttp.ClientSession, base: str, batch_id: str) -> str | None:
+    try:
+        async with session.get(f"{base}/api/v1/queue/default/i/{batch_id}") as resp:
+            if resp.ok:
+                data = await resp.json(content_type=None)
+                for node_result in data.get("session", {}).get("results", {}).values():
+                    if node_result.get("type") == "image_output":
+                        return node_result["image"]["image_name"]
+    except Exception as e:
+        logger.error(f"Item fetch error: {e}")
+
+    try:
+        async with session.get(
+            f"{base}/api/v1/images/",
+            params={"limit": 1, "order_by": "created_at", "direction": "DESC"},
+        ) as resp:
+            if resp.ok:
+                items = (await resp.json(content_type=None)).get("items", [])
+                if items:
+                    return items[0]["image_name"]
+    except Exception as e:
+        logger.error(f"Image list fallback error: {e}")
+
+    return None
+
+
+async def _fetch_image_bytes(session: aiohttp.ClientSession, base: str, name: str) -> bytes | None:
+    """Try the full image endpoint, then fall back to thumbnail."""
+    urls = [
+        f"{base}/api/v1/images/i/{name}/full",
+        f"{base}/api/v1/images/i/{name}/thumbnail",
+    ]
+    for url in urls:
+        try:
+            async with session.get(url) as resp:
+                ct   = resp.headers.get("Content-Type", "")
+                data = await resp.read()
+                logger.info(f"Image fetch {url}: status={resp.status} content-type={ct} size={len(data)}")
+                if "html" not in ct and len(data) >= _MIN_IMAGE_BYTES:
+                    return data
+        except Exception as e:
+            logger.error(f"InvokeAI image fetch error ({url}): {e}")
+    return None
+
+
 class GarfAI:
     def __init__(self):
         self.baseurl   = config.BASE_URL
-        self.openaikey = config.OPENAI_TOKEN
         self.sysprompt = config.SYSTEM_PROMPT
         self.txtmodel  = config.TXT_MODEL
         self.imgmodel  = config.IMG_MODEL
+
+        self._oai = AsyncOpenAI(
+            api_key=config.OPENAI_TOKEN,
+            base_url=config.BASE_URL,
+        )
         self.image_request_queue = asyncio.Queue()
 
     async def garfpic(self, ctx, prompt):
         await self.image_request_queue.put({"ctx": ctx, "prompt": prompt})
 
-    async def generate_image(self, prompt):
-        client = AsyncOpenAI(api_key=self.openaikey)
+    async def generate_image(self, session: aiohttp.ClientSession, prompt: str) -> bytes | str:
+        base = INVOKEAI_BASE
+
         try:
-            response = await client.images.generate(
-                model=self.imgmodel, prompt=prompt, n=1, size="1024x1024"
-            )
-        except openai.BadRequestError as e:
-            return f"`GarfBot Error: ({e.status_code}) - Your request was rejected as a result of our safety system.`"
-        except openai.InternalServerError as e:
-            logger.error(e)
-            return f"`GarfBot Error: ({e.status_code}) - Monday`"
+            async with session.post(
+                f"{base}/api/v1/queue/default/enqueue_batch",
+                json={"batch": {"graph": _build_graph(prompt), "runs": 1}},
+            ) as resp:
+                if not resp.ok:
+                    text = await resp.text()
+                    logger.error(f"InvokeAI enqueue failed {resp.status}: {text}")
+                    return "`GarfBot Error: InvokeAI rejected the request`"
+                data     = await resp.json(content_type=None)
+                batch_id = data["batch"]["batch_id"]
         except Exception as e:
-            logger.error(e)
-            return "`GarfBot Error: Lasagna`"
-        data = getattr(response, "data", None)
-        if not data:
-            logger.error("No data in response")
-            return "`GarfBot Error: No images generated`"
+            logger.error(f"InvokeAI enqueue error: {e}")
+            return "`GarfBot Error: Couldn't reach InvokeAI`"
 
-        first_image = data[0] if len(data) > 0 else None
-        if not first_image:
-            logger.error("No image in response data")
-            return "`GarfBot Error: No images generated`"
+        logger.info(f"InvokeAI batch queued: {batch_id}")
 
-        image_url = getattr(first_image, "url", None)
-        if not image_url:
-            logger.error("No URL in image response")
-            return "`GarfBot Error: No image URL returned`"
+        if not await _poll_batch(session, base, batch_id):
+            return "`GarfBot Error: InvokeAI generation failed or timed out`"
 
-        return image_url
+        image_name = await _get_image_name(session, base, batch_id)
+        if not image_name:
+            return "`GarfBot Error: Could not resolve image name`"
+
+        logger.info(f"Got image: {image_name}")
+
+        data = await _fetch_image_bytes(session, base, image_name)
+        if data:
+            return data
+
+        logger.error("All image download attempts failed")
+        return "`GarfBot Error: Odie`"
 
     async def process_image_requests(self):
-        async with aiohttp.ClientSession() as session:
+        async with aiohttp.ClientSession(headers={"Accept": "application/json"}) as session:
             while True:
                 request = await self.image_request_queue.get()
                 ctx    = request["ctx"]
                 prompt = request["prompt"]
-                image_url = await self.generate_image(prompt)
-                if image_url and "GarfBot Error" not in image_url:
-                    logger.info("Downloading & sending image...")
-                    async with session.get(image_url) as resp:
-                        if resp.status == 200:
-                            image_data = await resp.read()
-                            image = io.BytesIO(image_data)
-                            image.seek(0)
+
+                result = await self.generate_image(session, prompt)
+
+                if isinstance(result, bytes):
+                    logger.info("Sending image...")
+                    image     = io.BytesIO(result)
                     timestamp = ctx.message.created_at.strftime("%Y%m%d%H%M%S")
                     filename  = f"{timestamp}_generated_image.png"
-                            sendfile = discord.File(fp=image, filename=filename)
                     try:
-                                await ctx.send(file=sendfile)
+                        await ctx.reply(file=discord.File(fp=image, filename=filename))
                     except Exception as e:
                         logger.error(e)
                 else:
-                            await ctx.send("`GarfBot Error: Odie`")
-                else:
-                    await ctx.send(image_url)
+                    await ctx.reply(result)
+
                 self.image_request_queue.task_done()
                 await asyncio.sleep(2)
 
-    async def generate_chat(self, question):
+    async def generate_chat(self, question: str) -> str:
         try:
-            client = AsyncOpenAI(
-                api_key=self.openaikey,
-                base_url=self.baseurl
-            )
-            response = await client.chat.completions.create(
+            response = await self._oai.chat.completions.create(
                 model=self.txtmodel,
                 messages=[
-                    {
-                        "role": "system",
-                        "content": self.sysprompt,
-                    },
+                    {"role": "system", "content": self.sysprompt},
                     {"role": "user",   "content": question},
                 ],
                 max_tokens=400,
                 temperature=1.2,
             )
-            answer = str(response.choices[0].message.content)
+            answer = response.choices[0].message.content
             return answer.replace("an AI language model", "a cartoon animal")
         except openai.BadRequestError as e:
             logger.error(e)
@@ -110,12 +228,15 @@ class GarfAI:
             logger.error(e)
             return "`GarfBot Error: Lasagna`"
 
-    async def wikisum(self, query):
+    async def wikisum(self, query: str) -> str:
         try:
             summary = wikipedia.summary(query)
-            garfsum = await self.generate_chat(
-                f"Please summarize in your own words: {summary}"
-            )
-            return garfsum
+            return await self.generate_chat(f"Please summarize in your own words: {summary}")
+        except wikipedia.exceptions.DisambiguationError as e:
+            options = ", ".join(e.options[:3])
+            return f"`GarfBot Error: Ambiguous query — did you mean: {options}?`"
+        except wikipedia.exceptions.PageError:
+            return "`GarfBot Error: No Wikipedia page found for that query`"
         except Exception as e:
-            return e
+            logger.error(e)
+            return f"`GarfBot Error: {e}`"