Spaces:
Running
Running
Georgii Savin
commited on
feat: use more precise prompts for music and image generation
Browse files- src/agent/image_agent.py +15 -1
- src/agent/llm_graph.py +5 -4
- src/agent/music_agent.py +18 -4
- src/agent/prompts.py +16 -0
- src/agent/runner.py +1 -1
src/agent/image_agent.py
CHANGED
@@ -2,6 +2,8 @@ from pydantic import BaseModel, Field
|
|
2 |
from typing import Literal, Optional
|
3 |
from agent.llm import create_light_llm
|
4 |
from langchain_core.messages import SystemMessage, HumanMessage
|
|
|
|
|
5 |
import logging
|
6 |
|
7 |
logger = logging.getLogger(__name__)
|
@@ -66,12 +68,24 @@ class ChangeScene(BaseModel):
|
|
66 |
|
67 |
image_prompt_generator_llm = create_light_llm(0.1).with_structured_output(ChangeScene)
|
68 |
|
69 |
-
async def generate_image_prompt(
|
70 |
"""
|
71 |
Generates a detailed image prompt string based on a scene description.
|
72 |
This prompt is intended for use with an AI image generation model.
|
73 |
"""
|
74 |
logger.info(f"Generating image prompt for the current scene: {request_id}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
response = await image_prompt_generator_llm.ainvoke(
|
76 |
[
|
77 |
SystemMessage(content=IMAGE_GENERATION_SYSTEM_PROMPT),
|
|
|
2 |
from typing import Literal, Optional
|
3 |
from agent.llm import create_light_llm
|
4 |
from langchain_core.messages import SystemMessage, HumanMessage
|
5 |
+
from agent.state import get_user_state, set_user_state
|
6 |
+
from agent.prompts import GAME_STATE_PROMPT
|
7 |
import logging
|
8 |
|
9 |
logger = logging.getLogger(__name__)
|
|
|
68 |
|
69 |
image_prompt_generator_llm = create_light_llm(0.1).with_structured_output(ChangeScene)
|
70 |
|
71 |
+
async def generate_image_prompt(user_hash: str, scene_description: str) -> ChangeScene:
|
72 |
"""
|
73 |
Generates a detailed image prompt string based on a scene description.
|
74 |
This prompt is intended for use with an AI image generation model.
|
75 |
"""
|
76 |
logger.info(f"Generating image prompt for the current scene: {request_id}")
|
77 |
+
|
78 |
+
state = get_user_state(user_hash)
|
79 |
+
scene = GAME_STATE_PROMPT.format(
|
80 |
+
lore=state.story_frame.lore,
|
81 |
+
goal=state.story_frame.goal,
|
82 |
+
milestones=",".join(m.id for m in state.story_frame.milestones),
|
83 |
+
endings=",".join(e.id for e in state.story_frame.endings),
|
84 |
+
history="; ".join(f"{c.scene_id}:{c.choice_text}" for c in state.user_choices),
|
85 |
+
last_choice=last_choice,
|
86 |
+
scene_description=scene_description
|
87 |
+
)
|
88 |
+
|
89 |
response = await image_prompt_generator_llm.ainvoke(
|
90 |
[
|
91 |
SystemMessage(content=IMAGE_GENERATION_SYSTEM_PROMPT),
|
src/agent/llm_graph.py
CHANGED
@@ -60,7 +60,7 @@ async def node_init_game(state: GraphState) -> GraphState:
|
|
60 |
first_scene = await generate_scene.ainvoke(
|
61 |
{"user_hash": state.user_hash, "last_choice": "start"}
|
62 |
)
|
63 |
-
change_scene = await generate_image_prompt(first_scene["description"]
|
64 |
logger.info(f"Change scene: {change_scene}")
|
65 |
await generate_scene_image.ainvoke(
|
66 |
{
|
@@ -94,7 +94,7 @@ async def node_player_step(state: GraphState) -> GraphState:
|
|
94 |
"last_choice": state.choice_text,
|
95 |
}
|
96 |
)
|
97 |
-
change_scene = await generate_image_prompt(next_scene["description"]
|
98 |
current_image = None
|
99 |
if scene_id and scene_id in user_state.scenes:
|
100 |
current_image = user_state.scenes[scene_id].image
|
@@ -107,8 +107,9 @@ async def node_player_step(state: GraphState) -> GraphState:
|
|
107 |
"change_scene": change_scene,
|
108 |
}
|
109 |
)
|
110 |
-
music_task =
|
111 |
-
await asyncio.gather(image_task, music_task)
|
|
|
112 |
state.scene = next_scene
|
113 |
return state
|
114 |
|
|
|
60 |
first_scene = await generate_scene.ainvoke(
|
61 |
{"user_hash": state.user_hash, "last_choice": "start"}
|
62 |
)
|
63 |
+
change_scene = await generate_image_prompt(state.user_hash, first_scene["description"])
|
64 |
logger.info(f"Change scene: {change_scene}")
|
65 |
await generate_scene_image.ainvoke(
|
66 |
{
|
|
|
94 |
"last_choice": state.choice_text,
|
95 |
}
|
96 |
)
|
97 |
+
change_scene = await generate_image_prompt(state.user_hash, next_scene["description"])
|
98 |
current_image = None
|
99 |
if scene_id and scene_id in user_state.scenes:
|
100 |
current_image = user_state.scenes[scene_id].image
|
|
|
107 |
"change_scene": change_scene,
|
108 |
}
|
109 |
)
|
110 |
+
music_task = generate_music_prompt(next_scene["description"])
|
111 |
+
_, music_prompt = await asyncio.gather(image_task, music_task)
|
112 |
+
asyncio.create_task(change_music_tone(state.user_hash, music_prompt))
|
113 |
state.scene = next_scene
|
114 |
return state
|
115 |
|
src/agent/music_agent.py
CHANGED
@@ -1,7 +1,9 @@
|
|
1 |
from pydantic import BaseModel
|
2 |
from agent.llm import create_light_llm
|
|
|
3 |
from langchain_core.messages import SystemMessage, HumanMessage
|
4 |
import logging
|
|
|
5 |
|
6 |
logger = logging.getLogger(__name__)
|
7 |
|
@@ -38,10 +40,22 @@ class MusicPrompt(BaseModel):
|
|
38 |
llm = create_light_llm(0.1).with_structured_output(MusicPrompt)
|
39 |
|
40 |
|
41 |
-
async def generate_music_prompt(
|
42 |
-
logger.info(f"Generating music prompt for the current scene: {
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
response = await llm.ainvoke(
|
44 |
-
[SystemMessage(content=system_prompt), HumanMessage(content=
|
45 |
)
|
46 |
-
logger.info(f"Music prompt generated
|
47 |
return response.prompt
|
|
|
1 |
from pydantic import BaseModel
|
2 |
from agent.llm import create_light_llm
|
3 |
+
from agent.prompts import GAME_STATE_PROMPT
|
4 |
from langchain_core.messages import SystemMessage, HumanMessage
|
5 |
import logging
|
6 |
+
from agent.state import get_user_state, set_user_state
|
7 |
|
8 |
logger = logging.getLogger(__name__)
|
9 |
|
|
|
40 |
llm = create_light_llm(0.1).with_structured_output(MusicPrompt)
|
41 |
|
42 |
|
43 |
+
async def generate_music_prompt(user_hash: str, scene_description: str) -> str:
|
44 |
+
logger.info(f"Generating music prompt for the current scene: {scene_description}")
|
45 |
+
|
46 |
+
state = get_user_state(user_hash)
|
47 |
+
scene = GAME_STATE_PROMPT.format(
|
48 |
+
lore=state.story_frame.lore,
|
49 |
+
goal=state.story_frame.goal,
|
50 |
+
milestones=",".join(m.id for m in state.story_frame.milestones),
|
51 |
+
endings=",".join(e.id for e in state.story_frame.endings),
|
52 |
+
history="; ".join(f"{c.scene_id}:{c.choice_text}" for c in state.user_choices),
|
53 |
+
last_choice=last_choice,
|
54 |
+
scene_description=scene_description
|
55 |
+
)
|
56 |
+
|
57 |
response = await llm.ainvoke(
|
58 |
+
[SystemMessage(content=system_prompt), HumanMessage(content=scene)]
|
59 |
)
|
60 |
+
logger.info(f"Music prompt generated")
|
61 |
return response.prompt
|
src/agent/prompts.py
CHANGED
@@ -13,6 +13,22 @@ Return ONLY a JSON object with:
|
|
13 |
Translate the lore, goal, milestones and endings to the language which is used in the game and setting description.
|
14 |
"""
|
15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
SCENE_PROMPT = """You are an AI agent for a visual novel game.
|
17 |
Your role is to process incoming data and generate the next scene description and choices.
|
18 |
Translate the scene description and choices into a language which is used in the Game Settings.
|
|
|
13 |
Translate the lore, goal, milestones and endings to the language which is used in the game and setting description.
|
14 |
"""
|
15 |
|
16 |
+
GAME_STATE_PROMPT = """
|
17 |
+
---Game Settings START---
|
18 |
+
Lore: {lore}
|
19 |
+
Goal: {goal}
|
20 |
+
Milestones: {milestones}
|
21 |
+
Endings: {endings}
|
22 |
+
---Game Settings END---
|
23 |
+
|
24 |
+
---User's actions START---
|
25 |
+
History: {history}
|
26 |
+
Last choice: {last_choice}
|
27 |
+
---User's actions END---
|
28 |
+
|
29 |
+
Game response to user's action: {scene_description}
|
30 |
+
"""
|
31 |
+
|
32 |
SCENE_PROMPT = """You are an AI agent for a visual novel game.
|
33 |
Your role is to process incoming data and generate the next scene description and choices.
|
34 |
Translate the scene description and choices into a language which is used in the Game Settings.
|
src/agent/runner.py
CHANGED
@@ -56,7 +56,7 @@ async def process_step(
|
|
56 |
ending_desc = ending_info.get("description") or ending_info.get(
|
57 |
"condition", ""
|
58 |
)
|
59 |
-
change_scene = await generate_image_prompt(
|
60 |
if change_scene.change_scene == "no_change":
|
61 |
change_scene.change_scene = "change_completely"
|
62 |
if not change_scene.scene_description:
|
|
|
56 |
ending_desc = ending_info.get("description") or ending_info.get(
|
57 |
"condition", ""
|
58 |
)
|
59 |
+
change_scene = await generate_image_prompt(user_hash, ending_desc)
|
60 |
if change_scene.change_scene == "no_change":
|
61 |
change_scene.change_scene = "change_completely"
|
62 |
if not change_scene.scene_description:
|